1
$\begingroup$

From wikipedia, a random variable $X$ is said to have the Poisson distribution if for $\lambda > 0$,

$ \Pr(X = k)= \frac{\lambda^k e^{-\lambda}}{k!}$

Why can't the value $\lambda = 0$ be admitted? Wouldn't this physically correspond to the situation where if you expect the rate of event occurrence to be zero, assign all probability to observing the zero outcome.

$\endgroup$
  • 1
    $\begingroup$ Is there any problem with the formula when $\lambda=0$? $\endgroup$ – whuber Jan 13 '16 at 6:04
  • $\begingroup$ Yes there is, but in this case would it not be convenient to define $0^0 = 1$? Put it in a different way, what does the sequence of Poisson distributions converge to as $\lambda \rightarrow 0$. $\endgroup$ – Alex Jan 13 '16 at 6:10
  • 3
    $\begingroup$ homeschoolmath.net/teaching/zero-exponent-proof.php $\endgroup$ – Christoph Hanck Jan 13 '16 at 7:29
5
$\begingroup$

In general Poisson distribution is defined as having parameter $\lambda > 0$. For $\lambda < 1$ the smaller $\lambda$ gets, the more mass is accumulated around $0$, i.e. $\lim_{\lambda \rightarrow 0} P(K=0|\lambda) = 1$ and $\lim_{\lambda \rightarrow 0} P(K=k|\lambda) = 0$ for $k>0$ (Said, 1958). This is shown on an ugly plot below, that illustrates Poisson pmf for $\lambda = 1/10^0, ..., 1/10^{20}$ and $\lambda = 0$ as a red line (notice that it is heavily zoomed).

enter image description here

As noted by you and @ChristophHanck, number to the zeroth power is one and common convention is that $0^0 = 1$. On another hand $0^x = 0$ for non-zero $x$'s. Dividing zero by non-zero gives you zero. So if $\lambda = 0$ Poisson pmf simplifies to

$$ \frac{0^k \times \text{(whatever)}}{ \text{(whatever)} } $$

so it is degenerate distribution with all point mass at zero (cf. here):

$$ f(k) = \begin{cases} 1 & \text{if }k=0, \\[6pt] 0 & \text {if }k>0.\end{cases} $$

Mean and variance for Poisson distribution are equal to $\lambda$ and in this case only zero has non-zero probability, so expected value is obvious and there is no variability (variance is zero). Also R does not have any problem with $\lambda$'s defined as non-negative (rather than positive) values:

> dpois(0:10, 1e-10)
 [1]  1.000000e+00  1.000000e-10  5.000000e-21  1.666667e-31  4.166667e-42  8.333333e-53
 [7]  1.388889e-63  1.984127e-74  2.480159e-85  2.755732e-96  2.755732e-107
> dpois(0:10, 0)
 [1] 1 0 0 0 0 0 0 0 0 0 0
> dpois(0:10, -1)
 [1] NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN
Warning message:
In dpois(0:10, -1) : NaNs produced

So the behavior of Poisson pmf at $\lambda=0$ is coherent with its behavior at limit, moreover there is no problem with calculating Poisson pmf form this value, but it yields a degenerate distribution.


Said, A.S. (1958). Some properties of the poisson distribution. AIChE Journal, 4(3), 290-292.

$\endgroup$
  • $\begingroup$ well, if R says it is alright then it must be alright. Question: do standard textbooks for the Poisson distribution define it at $\lambda = 0$? $\endgroup$ – Alex Jan 14 '16 at 4:42
  • 1
    $\begingroup$ @Alex no they generally don't because it's degenerate, but you can define it like this and it's sometimes done so (I provided one reference that mentions it and R does so). $\endgroup$ – Tim Jan 14 '16 at 6:05
  • $\begingroup$ @Alex I updated answer with this comment. $\endgroup$ – Tim Jan 14 '16 at 8:22

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.