I have the following question:

I have two types of battery, $1$ and $2$. Suppose the lifetime of $1$ is uniformly distributed on the interval $(0,3)$, battery $2$ uniformly distributed on the interval $(10,13)$. $p=90\%$ of my batteries are type 1, $1-p=10\%$ are type $2$. Battery $1$ costs £$1$ per battery, Battery $2$ costs £$5$ per battery. I have a device which runs on one battery until the battery runs out, then I replace it. At each renewal, I pick a battery at random.

  1. For a large time $t$, what is the approximate probability that my device is being powered by a Type $2$ battery at time $t$?
  2. Would you recommend a different mix of batteries? What's the long term average cost per unit time for your preferred mix of batteries?

The first part sort of reminds of the inspection paradox/waiting time paradox, but I don't really know what to do. I feel like it is to do with the limiting distributions of the age $A_t=T_{X_t+1}-t$ and excess lifetime $E_t=t-T_{X_t}$.


Consider the Markov renewal-reward process $\{(X_n, T_n) : n = 1,2,3,\ldots\}$ where

  • $\tau_n$ is the lifetime of the $n^{\mathrm{th}}$ battery, $n\geqslant 1$
  • $T_n=\sum_{k=1}^n \tau_k$, $n\geqslant 1$
  • $\{X_n\}$ is a Markov chain on $\{1,2\}$ with transitions according to the arrival process $\{T_n\}$,
  • $\{Y_n\}$ is the reward sequence, i.e. $Y_n = -B_n - 5(1-B_n)$ where $B_n:=\mathsf 1_{\{X_n=1\}}$.

Then $\tau_n=B_nU_n + (1-B_n)V_n$ where $\{U_n\}$ is an iid sequence of $\mathcal U(0,3)$ random variables and $\{V_n\}$ is an iid sequence of $\mathcal U(10,13)$ random variables. As the $\{X_n\}$ are iid with $$\mathbb P(X_n=1)=p=1-\mathbb P(X_n=2) $$ for all $n$, it follows that the interrival times $\tau_n$ are iid with mixture density $$f(t) = \frac9{10}\left(\frac13\cdot\mathsf 1_{(0,3)}(t)\right) + \frac1{10}\left(\frac13\cdot\mathsf 1_{(10,13)}(t)\right) = \frac3{10}\cdot\mathsf 1_{(0,3)}(t) + \frac1{30}\cdot\mathsf 1_{(0,3)}(t), $$ and the $\{Y_n\}$ are iid with distribution $$ \mathbb P(Y_n=j) = \begin{cases} \frac9{10}(-1)\mathbb P(B_n=1) = -\frac9{10},&j=1\\ \frac1{10}(-5)\mathbb P(B_n=2) = -\frac12,& j=2. \end{cases} $$ Question $(a)$ concerns the asymptotic distribution of $\{\mathcal B(t):t\geqslant 0\}$ where $\mathcal B(t) = X_{\sup\{n\ :\ T_n<t\}}$, i.e. $\mathcal B(t)=j$ iff the last renewal $T_n$ before time $t$ was of battery type $j$. If we define $Y_n=T_nB_n$ and $Z_n=T_n(1-B_n)$, that is, splitting the process into an alternating renewal process which changes states when the type of battery changes, then we may compute \begin{align} \mathbb E[Y_n] &= \mathbb E[\tau_n\mid B_n=1] + \mathbb P(B_n=1)\mathbb E[Y_n]\\ &= \frac32 + \frac9{10}\mathbb E[Y_n]\\ \end{align} so that $\mathbb E[Y_n] = 15$, and similarly $$\mathbb E[Z_n] = \frac{23}2 + \frac1{10}\mathbb E[Z_n] $$ so that $\mathbb E[Z_n] = \frac{115}9$. It follows from the renewal-reward theorem that $$\lim_{t\to\infty} \mathcal B(t) = \frac{\mathbb E[Z_n]}{\mathbb E[Z_n]+\mathbb E[Y_n]} = \frac{\frac{115}9}{\frac{115}9+\frac{23}2}=\frac{23}{50}. $$ Note that this is the same as if we apply a similar logic to a single renewal epoch: $$\frac{\frac1{10}\cdot \frac{23}2}{\frac1{10}\cdot \frac{23}2 +\frac9{10}\frac32}=\frac{23}{50}. $$

As for $(b)$, we compute the expected interrenewal times $$\mathbb E[\tau_n] = \frac3{10}\int_0^3 t\ \mathsf dt + \frac1{30}\int_{10}^{13} t\ \mathsf d t = \frac52$$ and expected reward per epoch $$\mathbb E[Y_n] = \frac9{10}(-1) + \frac1{10}(-5) = -\frac75. $$ (The reward is negative because it is a cost.) Let $N(t)=\sum_{n=1}^\infty \mathsf 1_{(0,t]}(\tau_n)$ be the number of arrivals up to time $t$, and $W(t)=\sum_{n=1}^{N(t)}Y_n$ be the total award accumulated up to time $t$. Then by the renewal-reward theorem, the asymptotic long run average reward per unit time is $$\lim_{t\to\infty}\frac1t W(t) = \frac{\mathbb E[Y_1]}{\mathbb E[T_1]} = \frac{-7/5}{5/2} = -\frac{14}{25}.$$ If we are allowed to change $p$, then the above becoves $$R(p):=\frac{10-8p}{20p-23}, $$ which has strictly negative derivative on $(0,1)$, so in order to minimize cost, we should choose $p$, i.e. choose battery type 2.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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