0
$\begingroup$

For an M/M/1 queueing system, distribution of the time ($t_e$) between two consecutive events (either arrival or departure) can be derived as follows with the independent assumption,

$$F(t_e\ge t)=F(t_a\ge t)\cdot F(t_d\ge t)=e^{-\lambda t}e^{-\mu t}=e^{-(\lambda+\mu)t}$$

where $t_a$ and $t_d$ are the time between two consecutive arrivals and departures respectively.

From the complementary CDF (CCDF) derived above, we can say that $t_e$ is also exponentially distributed (with mean $1/(\lambda+\mu)$).

However, if we derive the result using CDF rather than CCDF, the outcome is,

$$F(t_e\le t)=F(t_a\le t)\cdot F(t_d\le t)\\=(1-e^{-\lambda t})(1-e^{-\mu t})=1-e^{-\lambda t}-e^{-\mu t}+e^{-(\lambda+\mu)t}\\<1-e^{-(\lambda+\mu)t}=1-F(t_e\ge t)$$

From this point of view, it seems we cannot say $t_e$ is exponentially distributed.

I am quite puzzled about this, could anyone shed some light upon this? Thanks.

$\endgroup$
1
$\begingroup$

Your first attempt uses the claim that $$ \{ t_e \ge t \} = \{ t_a \ge t \} \cap \{ t_d \ge t \}. \tag{$\dagger$} $$ (Here the "$\{...\}$" notation just means "the event that [...] happens".)

In your second attempt, you look at $\{t_e \le t\}$. Since everything is with exponential times, "$\le$" or "<" doesn't matter; let's look at $\{t_e < t\}$. You then claim that $$ \{ t_e < t \} = \{ t_a < t \} \cap \{ t_d < t \}. \tag{$\ddagger$}$$

However, these two displays, $(\dagger, \ddagger)$, are incompatible. Note that $\{t_e \ge t\} = \{t_e < t\}^c$. (Here $\{...\}^c$ means the complementary event.) But, for general sets/events $A$ and $B$, $$ \text{if}\quad C = A \cap B \quad\text{then}\quad C^c = A^c \cup B^c. $$ Hence, applying this in your situation, assuming that $(\dagger)$ is the correct statement, we get $$ \{t_e < t\} = \{t_e \ge t\}^c = \bigl( \{t_a \ge t\} \cap \{t_d \ge t\} \bigr)^c = \{t_a < t\} \cup \{t_d < t\}, $$ which you then need to deal with -- basically, you deal with this by doing your first computation for the different parts, and using the independence of $\{t_a \ge t\}$ and $\{t_d \ge t\}$ (which you assumed in your first calculation).

$\endgroup$
  • $\begingroup$ Thanks, it points out where I went wrong. Since the r.v. $t_e$ is the time till the next event, therefore $t_e\ge t$ means $t_a\ge t$ as well as $t_d\ge t$. However, for $t_e\le t$, at least one of $t_a\le t$ and $t_d\le t$ happens will guarantee $t_e\le t$. The probability, therefore, is equal to $1-P(t_a\ge t, t_d\ge t)=1-F(t_e\ge t)$. I see you applied de Morgan's law, but $C^c$ is not equal to $(A\cup B)\backslash(A\cap B)$ if $(A\cup B)$ is not the universal set. It's supposed to be $S\backslash(A\cap B)$. $\endgroup$ – Guoyang Qin Feb 12 at 19:14
  • 1
    $\begingroup$ Oops, thank you for pointing out the mistake! Let me just correct that... $\endgroup$ – Sam T Feb 12 at 20:52

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.