If $X$ and $Y$ are independent geometric random variables, $X \sim G(p)$ and $Y \sim G(q)$ then if $Z = X+Y$ I need to show that:

$$P[Z=z] = \frac{pq}{p-q}[(1-q)^{z-1}-(1-p)^{z-1}]$$

My attempt:
\begin{align} P[Z=z] &= \sum_{k=1}^z P[X=k]P[Y=z-k] && \text{(not sure of my summation limits here)} \\ &= \sum_{k=1}^z p(1-p)^{k-1} \cdot q(1-q)^{z-k-1} \\ &= \frac{pq}{(1-p)}\cdot \frac{(1-q)^z}{(1-q)}\cdot\sum_{k=1}^z (\frac{1-p}{1-q})^k \\ &= \frac{pq}{(1-p)}\cdot \frac{(1-q)^z}{(1-q)} \cdot \frac{1-(\frac{1-p}{1-q})^k}{1-\frac{1-p}{1-q}} \\ &= \frac{pq}{(1-p)}\cdot \frac{(1-q)^z}{(p-q)} \left[1 -(\frac{1-p}{1-q})^k\right] \\ &= \frac{pq}{p-q}\left[\frac{(1-q)^z}{1-p} - \frac{(1-p)^z}{1-p}\right] \end{align}

The result is pretty close to the answer. The only discrepancy is the first expression in the parentheses, $\frac{(1-q)^z}{1-p}$ which should be $\frac{(1-q)^z}{1-q}$.

Could someone please have a look at my working and show me where I have gone wrong? Or perhaps a neater calculation? Thanks!

I did check similar questions answered here, but couldn't find anything relevant to my problem.
Is the sum of two independent geometric random variables with the same success probability a geometric random variable?
How to compute the sum of random variables of geometric distribution


I think you made a mistake in the calculation. First geometric random variable is at least 1, hence $z\ge 2$.

\begin{align*} P[Z=z] &= \sum_{k=1}^{z-1} P[X=k]P[Y=z-k]\\ &=\sum_{k=1}^{z-1} p(1-p)^{k-1} \cdot q(1-q)^{z-k-1}\\ &=\frac{pq}{(1-p)} (1-q)^{z-1}\sum_{k=1}^{z-1} (\frac{1-p}{1-q})^k \\ &=\frac{pq}{(1-p)} (1-q)^{z-1} \cdot \frac{1-p}{1-q}\frac{1-(\frac{1-p}{1-q})^{z-1}}{1-\frac{1-p}{1-q}}\\ &=\frac{pq}{p-q} (1-q)^{z-1} \cdot [1 -(\frac{1-p}{1-q})^{z-1}]\\ &=\frac{pq}{p-q}[(1-q)^{z-1} - (1-p)^{z-1}], \end{align*} which is the desired result.

  • $\begingroup$ Thanks for the helpful answer. You lost me in the fourth line: $=\frac{pq}{(1-p)} (1-q)^{z-1} \cdot \frac{1-p}{1-q}\frac{1-(\frac{1-p}{1-q})^{z-1}}{1-\frac{1-p}{1-q}}$ How did you get the $\frac{1-p}{1-q}$ part before the summation? $\endgroup$ – abruzzi26 Feb 8 '17 at 15:22
  • $\begingroup$ @abruzzi26 This follows for the formula for the summation of geometric series with finite terms $\endgroup$ – John Feb 8 '17 at 15:26
  • $\begingroup$ Ah, yes. I got confused there for a sec. Thanks again $\endgroup$ – abruzzi26 Feb 8 '17 at 16:55

I like using the probability generating function for this: you know the pgf for the sum is the product of the pgfs for the two random variables. So

$$p_X(t)p_Y(t)={pqt^2\over (1-t(1-p))(1-t(1-q))}=$$

Applying partial fractions to just the reciprocal terms and setting consecutively $t=(1-p)^{-1}$ then $t=(1-q)^{-1}$ gives

$$A(1-t(1-q))+B(1-t(1-p))=1\iff A={1-p\over q-p}, B = -{1-q\over q-p}$$

leaving us with

$$p_{X+Y}(t) = {pqt^2\over q-p}\left({(1-p)\over 1-t(1-p)}-{1-q\over 1-t(1-q)}\right)$$ $$=\sum_{k=0}^\infty {pq\over q-p}((1-p)^{k+1}-(1-q)^{k+1})t^{k+2}$$

and reindexing appropriately gives

$$p_{X+Y}(t) = \sum_{i=2}^\infty {pq\over p-q}\Big((1-q)^{k-1}-(1-p)^{k-1}\Big)t^k$$

as desired.


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.