# Density function of a sum of 75 discrete random variables

I need help with find the density function of $$S$$ if

$$S=\sum_1^{75}X_j,$$

where $$X_j=I_jB_j$$, and $$I_1,\ldots, I_{75},B_1,\ldots, B_{75}$$ are independent,

$$Pr(I_j=1)=0.01 , \ Pr(I_j=0)=0.99 \ \ \mbox{ for all } j,$$

$$Pr(B_j=50)=0.7, Pr(B_j=100)=0.3 \ \ \mbox{ for } 1 \leq j \leq 50$$ and $$Pr(B_j=75)=0.7, Pr(B_j=150)=0.3 \ \ \mbox{ for } 51\leq j \leq 75$$

I think that maybe I have to calculate convolutions, but that is a tedious job and I don't know how could I progam that in python, for example. Is there an easier way to calculate the density function of S?

Thanks for any help.

• You have to show exact form of this pdf, or just program it in python? – vermator Mar 22 at 9:29
• I have to show a exact form of this pdf, but I can program it in python. – Mainnet Mar 22 at 9:35
• Are $\ I_j, B_j\$ all indepenent? – lonza leggiera Mar 22 at 10:54
• yes, they are independent – Mainnet Mar 22 at 10:59

I'm not sure what do you mean by density function in this context, because I think about the probability density function which is for continuous variables and here your outcome $$X_j$$ is discrete, therefore, the sum $$S$$ of $$X_j$$ for $$1 \leq j \leq 75$$ will also be discrete.

However, I can give some suggestions about the probability mass function, which is a similar concept but for discrete variables.

We can begin by looking for the probability mass function of $$X_j$$. I think it's better to start by analysing the problem in two parts, $$1$$ to $$50$$ and then $$51$$ to $$75$$. Then compute the possible products $$I_j B_j$$. In this case we will have:

For $$1 \leq j \leq 50$$: \begin{align} & I_j B_j = 1 \times 50 & \textrm{with} \quad Pr = 0.01 \times 0.7 \\ & I_j B_j = 1 \times 100 & \textrm{with} \quad Pr = 0.01 \times 0.3 \\ & I_j B_j = 0 \times 50 & \textrm{with} \quad Pr = 0.99 \times 0.7 \\ & I_j B_j = 0 \times 100 & \textrm{with} \quad Pr = 0.99 \times 0.3 \end{align} And for $$51 \leq j \leq 75$$: \begin{align} & I_j B_j = 1 \times 75 & \textrm{with} \quad Pr = 0.01 \times 0.7 \\ & I_j B_j = 1 \times 150 & \textrm{with} \quad Pr = 0.01 \times 0.3 \\ & I_j B_j = 0 \times 75 & \textrm{with} \quad Pr = 0.99 \times 0.7 \\ & I_j B_j = 0 \times 150 & \textrm{with} \quad Pr = 0.99 \times 0.3 \end{align}

Can you construct the probability mass function $$P(X_j=x)$$ from this data?

Edit: Just for fun, here is $$P(X_j=x)$$: $$P(X_j=x) \begin{cases} 0.990 & \textrm{if} & x=0 \\ 0.007 & \textrm{if} & x=50 \wedge 1 \leq j \leq 50 \\ 0.007 & \textrm{if} & x=75 \wedge 51 \leq j \leq 75 \\ 0.003 & \textrm{if} & x=100 \wedge 1 \leq j \leq 50 \\ 0.003 & \textrm{if} & x=150 \wedge 51 \leq j \leq 75 \\ 0 & & \textrm{otherwise} \end{cases}$$

And we can split the sums in two parts. Since the distribution in each part is multinomial, we get (if I'm not mistaken): $$\sum_{j=1}^{50} X_j = \sum_{j=0}^{50} \sum_{k=0}^{50-j} (50j + 100k) \times \frac{50!}{j! \cdot k! \cdot (50-j-k)!} \times 0.007^j \times 0.003^k \times 0.990^{50-j}$$ $$\sum_{j=51}^{75} X_j = \sum_{j=0}^{25} \sum_{k=0}^{25-j} (75j + 150k) \times \frac{25!}{j! \cdot k! \cdot (25-j-k)!} \times 0.007^j \times 0.003^k \times 0.990^{50-j}$$

• Thank you. I used python to calculate the coefficients of the probability generating function of $S$ and so get the probability $P(S=s)$. – Mainnet Mar 24 at 14:10
• You're welcome, @Minysh. Just for fun, I added \$P(X_j=x). :) – Ertxiem Mar 24 at 20:46