โ€œํ™•๋ฅ ๊ณผ ํ†ต๊ณ„(MATH230)โ€ ์ˆ˜์—…์—์„œ ๋ฐฐ์šด ๊ฒƒ๊ณผ ๊ณต๋ถ€ํ•œ ๊ฒƒ์„ ์ •๋ฆฌํ•œ ํฌ์ŠคํŠธ์ž…๋‹ˆ๋‹ค. ์ „์ฒด ํฌ์ŠคํŠธ๋Š” Probability and Statistics์—์„œ ํ™•์ธํ•˜์‹ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค ๐ŸŽฒ

9 minute read

โ€œํ™•๋ฅ ๊ณผ ํ†ต๊ณ„(MATH230)โ€ ์ˆ˜์—…์—์„œ ๋ฐฐ์šด ๊ฒƒ๊ณผ ๊ณต๋ถ€ํ•œ ๊ฒƒ์„ ์ •๋ฆฌํ•œ ํฌ์ŠคํŠธ์ž…๋‹ˆ๋‹ค. ์ „์ฒด ํฌ์ŠคํŠธ๋Š” Probability and Statistics์—์„œ ํ™•์ธํ•˜์‹ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค ๐ŸŽฒ

์ด๋ฒˆ ํฌ์ŠคํŠธ์—์„œ๋Š” ํฌ์•„์†ก ํ”„๋กœ์„ธ์Šค์™€๋„ ๊ด€๋ จ์žˆ๋Š” <Exponential Distribution>์— ๋Œ€ํ•ด ์‚ดํŽด๋ณธ๋‹ค!

Exponential Distribution

๋จผ์ € ๋ถ„ํฌ์— ๋Œ€ํ•œ ์‹์„ ๋จผ์ € ์ œ์‹œํ•˜๊ณ , ๊ทธ ์ƒํ™ฉ๊ณผ ์˜๋ฏธ๋ฅผ ์‚ดํŽด๋ณด์ž.

Definition. Exponential Distribution

Let $\lambda >0$, we say that $X$ has an <exponential distribution> with parameter $\lambda$ if it has pdf $f(x)$ as

\[f(x) = \begin{cases} \lambda e^{-\lambda x} & \text{for} \; x > 0\\ \quad 0 & \text{else} \end{cases}\]

, and we denote such RV $X$ as $X \sim \text{EXP}(\lambda)$.

Remark.

1. The cdf of $X$ is given by

\[\begin{aligned} P(X \le x) &= 1 - P(X > x) \\ &= 1 - e^{-\lambda x} \quad \text{for} \; x >0 \end{aligned}\]

์ด๋•Œ ์œ„์˜ cdf ์‹์—์„œ ์ฃผ๋ชฉํ•  ์ ์€ tail probability์ธ $P(X > x)$์ด๋‹ค. <Exponential Distribution>์˜ ๊ฒฝ์šฐ, $P(X > x)$๊ฐ€ $P(X > x) = e^{-\lambda x}$๋กœ ๊ณ„์‚ฐ๋จ์— ์ฃผ๋ชฉํ•˜์ž. EXP์— ๋Œ€ํ•œ ํ•ด์„ค์€ ์—ฌ๊ธฐ์„œ๋ถ€ํ„ฐ ์‹œ์ž‘๋œ๋‹ค.

EXP๋Š” <Poisson Process>์˜ ์ƒํ™ฉ์—์„œ๋ถ€ํ„ฐ ๋น„๋กฏ๋œ๋‹ค. ๋จผ์ € <Poisson Distribution>์€ $X$: [์–ด๋–ค ์‚ฌ๊ฑด์ด ๋ฐœ์ƒํ•˜๋Š” ํšŸ์ˆ˜]๋ฅผ ๋Œ€ํ‘œํ•˜๋ฉฐ, ๊ทธ ํ•จ์ˆ˜๋Š” ์•„๋ž˜์™€ ๊ฐ™๋‹ค.

\[f(x) = e^{-\lambda} \frac{\lambda^x}{x!}\]

๋งŒ์•ฝ, <Poisson Process> $\{ N(t) \}$์— ๋Œ€ํ•ด ์ƒ๊ฐํ•œ๋‹ค๋ฉด, $t$ ์‹œ๊ฐ„๊นŒ์ง€ ๋„์ฐฉํ•œ ์‚ฌ๊ฑด์˜ ์ˆซ์ž์ธ $N(t)$๋Š” ํฌ์•„์†ก ๋ถ„ํฌ $N(t) \sim \text{POI}(\lambda t)$๋ฅผ ๋”ฐ๋ฅธ๋‹ค.

์ด๋Ÿฐ ์ƒํ™ฉ์„ ์ƒ๊ฐํ•ด๋ณด์ž. โ€œ์–ด๋–ค ์‚ฌ๊ฑด์ด ์ฒ˜์Œ์œผ๋กœ ๋ฐœ์ƒํ•˜๊ธฐ ๊นŒ์ง€ ๊ฑธ๋ฆฐ ์‹œ๊ฐ„โ€์„ RV $T$๋ผ๊ณ  ํ•ด๋ณด์ž. ์šฐ๋ฆฌ๊ฐ€ $P(T > t)$ ์ฆ‰, ์–ด๋–ค ์‚ฌ๊ฑด์ด $T=t$ ์‹œ๊ฐ„ ์ดํ›„์— ๋ฐœ์ƒํ•  ํ™•๋ฅ ์„ ๊ตฌํ•œ๋‹ค๊ณ  ํ•ด๋ณด์ž. ์ด๊ฒƒ์„ Poisson Process์˜ ๊ด€์ ์—์„œ ํ•ด์„ํ•˜๋ฉด, $T=t$ ์‹œ๊ฐ„๊นŒ์ง€ ์–ด๋–ค ์‚ฌ๊ฑด๋„ ์ผ์–ด๋‚˜์ง€ ์•Š์€ ์ƒํƒœ, ์ฆ‰ $N(t) = 0$์ธ ์ƒํ™ฉ์ด๋‹ค. ์šฐ๋ฆฌ๋Š” $N(t)$์— ๋Œ€ํ•œ pdf๋ฅผ ์•Œ๊ธฐ ๋•Œ๋ฌธ์— ์ด๊ฒƒ์„ ํ™•๋ฅ ๋กœ ํ‘œํ˜„ํ•˜๋ฉด,

\[P(N(t) = 0) = e^{-\lambda t} \frac{(\lambda t)^0}{0!} = e^{-\lambda t}\]

๊ฐ€ ๋œ๋‹ค. ์ฆ‰, $P(T > t) = P(N(t) = 0) = e^{-\lambda t}$์ธ ์…ˆ์ด๋‹ค! ๐Ÿคฉ

์šฐ๋ฆฌ๊ฐ€ ๊ตฌํ•œ $P(T >t)$๋Š” $T$์— ๋Œ€ํ•œ tail probability์ด๋‹ค. ๊ทธ๋ž˜์„œ ์ด๊ฒƒ์„ cdf์˜ ํ˜•ํƒœ๋กœ ๋ฐ”๊ฟ”์ฃผ๋ฉด,

\[P(T \le t) = 1 - P(T > t) = 1 - e^{-\lambda t}\]

์ด๋‹ค. ์šฐ๋ฆฌ๊ฐ€ cdf๋ฅผ ์•Œ๊ณ  ์žˆ์œผ๋‹ˆ, ๋ฏธ๋ถ„์„ ํ†ตํ•ด pdf๋„ ๊ตฌํ•  ์ˆ˜ ์žˆ๋‹ค.

\[f(t) = \frac{d}{dt} P(T \le t) = \frac{d}{dt} (1 - e^{-\lambda t}) = \lambda e^{-\lambda t}\]

์ต์ˆ™ํ•œ ํ˜•ํƒœ์ด์ง€ ์•Š์€๊ฐ€?? ๋ฐ”๋กœ ์šฐ๋ฆฌ๊ฐ€ ์ •์˜ํ•œ <Exponential Distribution>์ด๋‹ค!! ๐Ÿ˜Ž

์ฆ‰, EXP๋Š” ์–ด๋–ค ์‚ฌ๊ฑด์ด ์ฒ˜์Œ์œผ๋กœ ์ผ์–ด๋‚  ์‹œ๊ฐ„์— ๋Œ€ํ•œ ํ™•๋ฅ  ๋ถ„ํฌ๋ผ๊ณ  ํ•  ์ˆ˜ ์žˆ๋‹ค!


<Exponential Distribution>์€ $\lambda$, $\beta$ ๋‘ ๊ฐ€์ง€ ํ˜•ํƒœ๋กœ ๊ธฐ์ˆ ํ•  ์ˆ˜ ์žˆ๋‹ค. ์ด๋•Œ, $\lambda$๋Š” Poisson Process์—์„œ ์œ ๋ž˜ํ•œ ๊ฒƒ์œผ๋กœ Time Unit(=interval) ๋‹น ๋ฐœ์ƒํ•˜๋Š” Event์˜ ํ‰๊ท ์ ์ธ ํšŸ์ˆ˜๋ฅผ ์˜๋ฏธํ•œ๋‹ค. EXP๋Š” $\beta$๋กœ๋„ ๋ชจ๋ธ๋งํ•  ์ˆ˜ ์žˆ๋Š”๋ฐ, ์ด๋•Œ $\beta$๋Š” $\lambda$์˜ ์—ญ์ˆ˜(reciprocol)์ด๋‹ค. ๋”ฐ๋ผ์„œ, $\beta$๋Š” ์ฒซ Event๊ฐ€ ๋ฐœ์ƒํ•˜๋Š”๊ฒŒ ๊ฑธ๋ฆฌ๋Š” ํ‰๊ท ์ ์ธ ์‹œ๊ฐ„์„ ์˜๋ฏธํ•œ๋‹ค.

\[X \sim \text{EXP}(\lambda) \iff f(x) = \lambda e^{-\lambda x}\]
  • $\lambda$๋Š” Unit time ๋™์•ˆ Event๊ฐ€ ์ผ์–ด๋‚  ํ‰๊ท  ํšŸ์ˆ˜๋ฅผ ์˜๋ฏธํ•œ๋‹ค.
\[X \sim \text{EXP}(\beta) \iff f(x) = \frac{1}{\beta} e^{-x/\beta}\]
  • $\lambda$์˜ ์—ญ์ˆ˜์ธ $\beta$๋Š” ํ•œ ๋ฒˆ์˜ Event๊ฐ€ ์ผ์–ด๋‚  ํ‰๊ท  ์‹œ๊ฐ„์„ ์˜๋ฏธํ•œ๋‹ค.

2. If $X \sim \text{EXP}(1)$, then $Y := \dfrac{X}{\lambda} \sim \text{EXP}(\lambda)$.

\[P(Y > y) = P(\frac{X}{\lambda} > y) = P(X > \lambda y) = e^{-\lambda y}\]

๋ณธ์ธ์€ ์œ„์˜ ์ƒํ™ฉ์„ (minute - second) ๋ณ€ํ™˜์„ ๋ฐ”ํƒ•์œผ๋กœ ์ดํ•ดํ–ˆ๋‹ค. ๋งŒ์•ฝ $X$๊ฐ€ ๋ถ„ ๋‹จ์œ„์—์„œ ์ฒ˜์Œ ๋„์ฐฉํ•˜๋Š” ๋ฒ„์Šค์˜ ์‹œ๊ฐ„์„ ๋ชจ๋ธ๋งํ•˜๊ณ , ๊ทธ ๋•Œ์˜ parameter๊ฐ€ $\lambda = 1$๋ผ๊ณ  ํ•˜์ž. ์šฐ๋ฆฌ๋Š” ์ด๊ฒƒ์„ ์ดˆ ๋‹จ์œ„์ธ $60X$๋กœ ๋ณ€ํ™˜ํ•  ์ˆ˜ ์žˆ๋‹ค. ์ด๋•Œ์˜ tail probability๋Š”

\[P(60X > x) = P(X > x/60) = e^{- x/60}\]

๋”ฐ๋ผ์„œ, $60X \sim \text{EXP}(1/60)$์ด ๋œ๋‹ค. ์ด๊ฒƒ์€ $60X$์—์„œ $\lambda$๊ฐ€ $\lambda = 1/60$์ด ๋จ์„ ์˜๋ฏธํ•œ๋‹ค. ์ด๋•Œ, $\lambda$๋Š” Poisson Process์˜ parameter๋กœ, Time Unit ๋‹น ๋„์ฐฉํ•˜๋Š” ๋ฒ„์Šค์˜ ์ˆ˜๋ฅผ ๋ชจ๋ธ๋งํ•œ๋‹ค. ๋”ฐ๋ผ์„œ 1์ดˆ ๋‹น ํ‰๊ท ์ ์œผ๋กœ 1/60 ๋Œ€์˜ ๋ฒ„์Šค๊ฐ€ ๋„์ฐฉํ•จ์„ ์˜๋ฏธํ•œ๋‹ค. ์ด๊ฒƒ์„ $\beta = 1 / \lambda$๋กœ ํ•ด์„ํ•˜๋ฉด, ๋ฒ„์Šค๊ฐ€ ํ•œ๋ฒˆ ๋„์ฐฉํ•˜๋Š” ์‹œ๊ฐ„์ด ํ‰๊ท ์ ์œผ๋กœ 60์ดˆ๊ฐ€ ๋จ์„ ์˜๋ฏธํ•œ๋‹ค!


3. (Memoryless Property) ์šฐ๋ฆฌ๋Š” ์•ž์„œ ์–ด๋–ค ์‚ฌ๊ฑด์ด ์ฒ˜์Œ์œผ๋กœ ๋ฐœํ–‰ํ•˜๋Š” ์‹œํ–‰ ํšŸ์ˆ˜ $X$๋ฅผ ๋ชจ๋ธ๋งํ•œ <Geometric Distribution>์„ ์‚ดํŽด๋ณธ ์ ์ด ์žˆ๋‹ค. ์–ด๋–ค ๋ถ„ํฌ๊ฐ€ <Memoryless Property>๋ฅผ ๊ฐ€์ง„๋‹ค๋ฉด, ์•„๋ž˜์˜ ์‹์„ ๋งŒ์กฑํ•œ๋‹ค.

\[P(X > a + t \mid X >a) = P(X > t)\]

EXP๊ฐ€ ์œ„์˜ Memoryless Property๋ฅผ ๊ฐ€์ง€๋Š”์ง€ ํ™•์ธํ•ด๋ณด์ž.

\[\begin{aligned} P(X > a + t \mid X > a) &= \frac{P(X > a + t)}{P(X > a)} \\ &= \frac{e^{-\lambda (a+t)x}}{e^{-\lambda a x}} \\ &= e^{-\lambda tx} = P(X > t) \end{aligned}\]

๋”ฐ๋ผ์„œ, EXP ์—ญ์‹œ Memoryless Property๋ฅผ ๊ฐ€์ง„๋‹ค!

<Geometric Distribution>์œผ๋กœ๋ถ€ํ„ฐ <Exponential Distribution>์„ ์œ ๋„ํ•ด๋ณผ ์ˆ˜๋„ ์žˆ๋Š”๋ฐ, ์•„๋ž˜์˜ ํŽผ์ณ๋ณด๊ธฐ์— ๊ธฐ์ˆ ํ•˜์˜€๋‹ค.

ํŽผ์ณ๋ณด๊ธฐ

Random Variable $X_n$์„ $1/n$์ดˆ๋งˆ๋‹ค ๋ฒ„์Šค๊ฐ€ ์™”๋Š”์ง€ ์•ˆ ์™”๋Š”์ง€ ํ™•์ธํ–ˆ์„ ๋•Œ, ๋ฒ„์Šค๊ฐ€ ์ฒ˜์Œ์˜ฌ ๋•Œ๊นŒ์ง€ ํ™•์ธํ•œ ํšŸ์ˆ˜๋ผ๊ณ  ํ•ด๋ณด์ž. ๋˜, $X$๋Š” ๋ฒ„์Šค๊ฐ€ ์ฒ˜์Œ์˜ฌ ๋•Œ๊นŒ์ง€ ๊ฑธ๋ฆฐ ์‹œ๊ฐ„์ด๋ผ๊ณ  ํ•œ๋‹ค๋ฉด, $X_n$์™€ $X$ ์‚ฌ์ด์—๋Š” ์•„๋ž˜์˜ ๋น„๋ก€์‹์ด ์„ฑ๋ฆฝํ•  ๊ฒƒ์ด๋‹ค.

\[X_n : X = 1 : \frac{1}{n}\]

๋˜, Geometric Distribution์„ ๋”ฐ๋ฅด๋Š” $X_n$์˜ parameter๋ฅผ $p$๋ผ๊ณ  ํ•˜์ž; $X_n \sim \text{Geo}(p)$, ๊ทธ๋Ÿฌ๋ฉด $E[X_n] = 1/p$๊ฐ€ ๋œ๋‹ค. ์ฆ‰, ํ‰๊ท ์ ์œผ๋กœ $1/p$๋ฒˆ ํ™•์ธํ•œ๋‹ค๋Š” ๋ง์ด๋‹ค. ์ด๊ฒƒ์„ ๋‹ค์‹œ $X$์˜ ๊ด€์ ์—์„œ ๊ธฐ์ˆ ํ•˜๋ฉด, ํ‰๊ท ์ ์œผ๋กœ $1/np$์ดˆ๊ฐ€ ๊ฑธ๋ฆฐ๋‹ค๋Š” ๋ง์ด๋‹ค. ์ฆ‰, $\beta = 1/np$๋ผ๋Š” ๋ง์ด๊ณ , $\lambda$๋กœ ํ‘œํ˜„ํ•˜๋ฉด, $\lambda = np$๋ผ๋Š” ๋ง์ด๋‹ค. ๋”ฐ๋ผ์„œ, $X_n \sim \text{Geo}\left( \frac{\lambda}{n} \right)$๊ฐ€ ๋œ๋‹ค.

์ด์— ๋”ฐ๋ผ, $X$์˜ tail probability $P(X > x)$๋Š”

\[\begin{aligned} P(X > x) &= P\left(\frac{X_n}{n} > x\right) \\ &= P(X_n > nx) \\ &= \left( 1 - \frac{\lambda}{n}\right)^{nx} \\ &= e^{-\lambda x} \quad \text{as } n \rightarrow \infty \end{aligned}\]

์ฆ‰, <Geometric Distribution>์—์„œ ๊ทนํ•œ์„ ์ทจํ•ด <Exponential Distribution>์„ ์œ ๋„ํ•  ์ˆ˜ ์žˆ๋‹ค!


Theorem.

Let $X \sim \text{EXP}(\lambda)$, then

  • $E[X] = \dfrac{1}{\lambda}$
  • $\text{Var}(X) = \dfrac{1}{\lambda^2}$

Proof.

Let $Y \sim \text{EXP}(1)$, then what are the mean and variacen of $Y$?

\[\begin{aligned} E[Y] &= \int^{\infty}_0 y \cdot e^{-y} \; dy = 1 \end{aligned}\]

Variance๋ฅผ ๊ตฌํ•ด๋ณด๋ฉด,

\[\begin{aligned} E[Y^2] = \int^{\infty}_0 y^2 \cdot e^{-y} \; dy = 2 \end{aligned}\]

๋”ฐ๋ผ์„œ, $\text{Var}(Y) = E[Y^2] - E[Y]^2 = 2 - 1^2 = 1$.

์ด์ œ, $X \sim \text{EXP}(\lambda)$๋ฅผ ์‚ดํŽด๋ณด์ž. ์šฐ๋ฆฌ๋Š” ์•ž์˜ <Remark 2>๋ฅผ ํ†ตํ•ด $X = \dfrac{Y}{\lambda}$์ž„์„ ์•Œ ์ˆ˜ ์žˆ๋‹ค. ๋”ฐ๋ผ์„œ,

\[E[X] = E\left[\frac{Y}{\lambda}\right] = \frac{1}{\lambda}\] \[\text{Var}(X) = \text{Var}\left( \frac{Y}{\lambda} \right) = \frac{1}{\lambda^2}\]

์š”์•ฝ.

  • ์–ด๋–ค ์‚ฌ๊ฑด์˜ ๋ฐœ์ƒ ํšŸ์ˆ˜๊ฐ€ ํฌ์•„์†ก ๋ถ„ํฌ๋ฅผ ๋”ฐ๋ฅธ๋‹ค๋ฉด, ์‚ฌ๊ฑด ์‚ฌ์ด์˜ ๋Œ€๊ธฐ ์‹œ๊ฐ„์€ ์ง€์ˆ˜ ๋ถ„ํฌ๋ฅผ ๋”ฐ๋ฅด๊ฒŒ ๋œ๋‹ค. (๋˜๋Š” ์ฒซ ์‚ฌ๊ฑด์ด ๋ฐœ์ƒํ•˜๋Š” ๋ฐ๊นŒ์ง€ ๊ฑธ๋ฆฌ๋Š” ์‹œ๊ฐ„์€ ์ง€์ˆ˜ ๋ถ„ํฌ๋ฅผ ๋”ฐ๋ฅธ๋‹ค.)
  • $\lambda$๋Š” Unit time ๋™์•ˆ Event๊ฐ€ ์ผ์–ด๋‚  ํ‰๊ท  ํšŸ์ˆ˜๋ฅผ ์˜๋ฏธํ•œ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ๊ทธ ์—ญ์ˆ˜์ธ $\beta$๋Š” ํ•œ ๋ฒˆ์˜ Event๊ฐ€ ๋ฐœ์ƒํ•˜๋Š” ๋ฐ ๊ฑธ๋ฆฌ๋Š” ํ‰๊ท  ์‹œ๊ฐ„์„ ์˜๋ฏธํ•œ๋‹ค.
  • <Exponential Distribution>์€ <Geometric Distribution>์˜ ๊ทนํ•œ ๋ฒ„์ „์ด๋‹ค. Geo์—์„œ trial์„ ์‹œํ–‰ํ•˜๋Š” ์‹œ๊ฐ„ ๊ฐ„๊ฒฉ $1/n$์ด 0์— ๊ฐ€๊นŒ์›Œ์งˆ ๋•Œ, Geo๊ฐ€ EXP๋ฅผ ๋”ฐ๋ฅด๊ฒŒ ๋˜๋Š” ๊ฒƒ์ด๋‹ค.

์ด์–ด์ง€๋Š” ํฌ์ŠคํŠธ์—์„œ๋Š” ์—ฐ์† ํ™•๋ฅ  ๋ถ„ํฌ์—์„œ <์ •๊ทœ ๋ถ„ํฌ>๋งŒํผ์ด๋‚˜ ์ค‘์š”ํ•œ ๋ถ„ํฌ์ธ <๊ฐ๋งˆ ๋ถ„ํฌ; Gamma Distribution>์— ๋Œ€ํ•ด ์‚ดํŽด๋ณธ๋‹ค! ๐Ÿคฉ

๐Ÿ‘‰ Gamma Distribution


References