I've been working on the following problem. I'm a bit confused about some of the specifics of how to arrive at the correct answer. I hope someone here could point me in the right direction:

A dart is tossed at a target with a Cartesian $(X_1,X_2)$ coordinate system centered in the middle of the bull's-eye. Assume $X_1$, the horizontal miss distance, and $X_2$, the vertical miss distance are independent, normal random variables, each with $\mu=0$, $\sigma^2=2$. The radial miss distance then is $R=\sqrt{X_1^2 + X_2^2}$, evaluate:

(a) $P(R>2)$

(b) The radius, $r$, of the circle centered at $(0,0)$, which has probability $0.5$ of containing $(X_1,X_2)$.

This is what I have so far:

Since $X_1$ and $X_2$ are normal variables, their square will follow the chi-squared distribution for random variables. For the "standard normal" random variable $Z$ (i.e. $\mu=0$ and $\sigma^2=1$), the random variable $ Y = Z^2$ would be:

$P(Y<t) = P(Z^2<t) = P(-\sqrt{t} < Z < \sqrt{t}) = 2N(\sqrt{t})-1$

where $N(z) = \int_{-\infty}^z{\frac{1}{\sqrt{2\pi}}}e^{-t^2/2}dt$ is the cumulative distribution function for the standard normal distribution

however, these distributions are not "standard normal", since the variance ($\sigma^2$) is equal to 2, not 1.

For normal distribution functions, the cumulative distribution function is:

$F_X(x) = N\left(\frac{x-\mu}{\sigma}\right) = \int_{-\infty}^{\frac{x-\mu}{\sigma}}{\frac{1}{\sqrt{2\pi}}}e^{-t^2/2}dt$

So, the square of the normal distribution function of the random variable $W=X^2$ should follow as:

$P(W<t) = P(X^2<t) = P(-\sqrt{t}<X<\sqrt{t}) = 2N\left(\frac{\sqrt{t}-\mu}{\sigma}\right)-1 = F_W(t)$

and the density function for these random variables would be:

$f_W(t) = \frac{d}{dt}F_W(t) = \frac{n\left(\frac{\sqrt{t}-\mu}{\sigma}\right)}{\sigma\sqrt{t}} = \frac{1}{\sigma\sqrt{2\pi t}}e^{-(\sqrt{t}-\mu)^2/2\sigma^2}$

and, since the average ($\mu$) = 0, this simplifies to:

$f_W(t) = \frac{1}{\sigma\sqrt{2\pi t}}e^{-t/2\sigma^2}$

which is, of course, a gamma distribution with parameters $\lambda = \frac{1}{2\sigma^2}$ and $n = \frac{1}{2}$

Now, up to this point, I'm fairly confident, it's the next part that I'm a bit confused by.

Now, we have the random variable $R = \sqrt{X_1^2+X_2^2}$, the probability that the radius hit is greater than 2 is:

$P(R>2) = 1-P(R<2) = 1-P(\sqrt{X_1^2+X_2^2} <2) = 1 - P(X_1^2+X_2^2 < 4) $

Now, if we define a random variable $K = X_1^2 + X_2^2$, this will be a sum of independent chi-squared variables, therefore it will be a gamma distribution with $\lambda = \frac{1}{2\sigma^2}$ and $ n = 2/2 = 1$:

$f_K(t) = \frac{1}{2\sigma^2} e^{-t/2\sigma^2}$

which is the Erlang probability ($P(K<t) = F_K(t) = 1-\sum_{k=0}^{r-1}{\frac{(\lambda t)^k}{k!}e^{-\lambda t}}$) law with r=1, therefore:

$P(K>t) = e^{-t/2\sigma^2} = e^{-t/4}$

so, for part (a):

$P(R>2) = P(X_1^2+X_2^2>4) = P(K>4) = e^{-4/4} = e^{-1}$

which is the correct answer according to the solutions manual

Part (b) asks to find the radius when the probability is 0.5 of hitting the target, therefore:

$P(R<r) = 1-e^{-r/4} = 0.5$

$e^{-r/4} = 0.5$

$\frac{-r}{4} = -ln(2)$

$r = ln(16)$

$r = 2.7725887...$

but since this is actually the radius squared, the value should be:

$radius = \sqrt{2.7725887} = 1.66511$

However, this is apparently incorrect. The correct answer for part (b) should be: 1.1774. This is the answer if you divide 2.7725887 by 2 before taking the square-root. I am very confused by why I've arrived at the wrong answer for part (b). If anyone could help, I would be very grateful!


I suspect you did too much work.

Let's make some observations. $$R = \sqrt{X_1^2+X_2^2} = \sqrt{\left(\sqrt 2\frac{X_1}{\sqrt 2}\right)^2+\left(\sqrt 2\frac{X_2}{\sqrt 2}\right)^2} = \sqrt 2\sqrt{\mathsf X_1^2+\mathsf X_2^2} = \sqrt 2\mathsf R$$ where $$\frac{X_1}{\sqrt 2} =\mathsf X_1\overset d= \mathsf X_2\sim N(0,1)$$ and hence $\mathsf R$ follows a standard Rayleigh distribution. So depending on the material covered, you might have overlooked this fact, or this is just another way to verify your answers.

For a) \begin{align*} P(R>2)&= P(\sqrt 2 \mathsf R >2)\\ &=1-P(\mathsf R \leq 2/\sqrt 2)\\ &=1-\left(1-\exp\left\{-\frac{1}{2}\left(\frac{2}{\sqrt 2}\right)^2\right\}\right)\tag 1\\ &=e^{-1}\\ &=0.3678794 \end{align*} where in $(1)$ I used the well known cdf of a Rayleigh.

Now for part b)

\begin{align*} P(R<r)&= \frac{1}{2}\\ P(\sqrt 2 \mathsf R <r) &= \frac 12\\ P\left(\mathsf R<\frac{r}{\sqrt 2}\right) &=\frac 12\\ 1-\exp\left\{-\frac{1}{2}\left(\frac{r}{\sqrt 2}\right)^2\right\} &=\frac{1}{2}\tag2\\ \exp\left\{-\frac{1}{2}\left(\frac{r}{\sqrt 2}\right)^2\right\}&=\frac{1}{2}\\ -\frac{1}{2}\frac{r^2}{2} &= -\log 2\\ r^2 &= 4\log 2 \end{align*} where in $(2)$ I used the well known cdf of a Rayleigh, and this gives that $r = 1.665109$.

  • $\begingroup$ Thank you for the answer, however part (b) still has the technically "incorrect" answer. Although, there is a possibility that the solutions in the back of the book are wrong (they have been before) $\endgroup$ – lstbl Feb 28 '16 at 14:48
  • 1
    $\begingroup$ @lstbl Let me put it bluntly. They are wrong. The three of us agree, and we used different methods. I highly doubt the solution is 'correct'. $\endgroup$ – Em. Feb 28 '16 at 14:55
  • $\begingroup$ @probablyme Nice solution, but perhaps you have a typo in part (a). The Rayleigh CDF is $1-e^{-x^2/2}$, therefore $1-(1-e^{-x^2/2})=e^{-x^2/2}=1/e$, in agreement with both myself, lstbl (and the book, which seems to be the least reliable of the currently available sources...;-) $\endgroup$ – Pierpaolo Vivo Feb 28 '16 at 15:28
  • $\begingroup$ ok (the numerical 0.6321206 is also to be edited ;-)) ) $\endgroup$ – Pierpaolo Vivo Feb 28 '16 at 15:33
  • $\begingroup$ @PierpaoloVivo Wow, sorry I was not paying attention. Thanks a lot. $\endgroup$ – Em. Feb 28 '16 at 15:39

Let us first compute in the quickest way the PDF of $R$ using Dirac delta. The joint PDF of $X_1$ and $X_2$ is $$ f_{X_1,X_2}(x_1,x_2)=\frac{e^{-x_1^2/4}}{\sqrt{4\pi}}\frac{e^{-x_2^2/4}}{\sqrt{4\pi}}\ , $$ therefore the PDF of $R$ is $$ f_R(r)=\int_{-\infty}^\infty dx\int_{-\infty}^\infty dy\frac{e^{-x_1^2/4}}{\sqrt{4\pi}}\frac{e^{-x_2^2/4}}{\sqrt{4\pi}}\delta\left(r-\sqrt{x^2+y^2}\right)\ , $$ which is easily handled in polar coordinates $x=\hat{r}\sin\theta$ and $y=\hat{r}\cos\theta$, yielding $$ f_R(r)=\frac{2\pi}{4\pi}\int_0^\infty d\hat{r}\ \hat{r} e^{-\hat{r}^2/4}\delta(r-\hat{r})=\frac{1}{2}r e^{-r^2/4}\ , $$ which is correctly normalized for $r\in (0,\infty)$. To compute part (A), we just need to integrate this PDF for $r>2$: $$ P(R>2)=\int_2^\infty dr f_R(r)=\int_2^\infty dr\frac{1}{2}r e^{-r^2/4}=1/\mathrm{e}\approx 0.3678.... $$ For part (B), we just need to integrate the radial PDF $f_R(r^\prime)$ up to an unspecified point $r$, and then solve a simple equation $$ P(0<R<r)=\int_0^r\ dr^\prime f_R(r^\prime)=1-e^{-\frac{r^2}{4}}\ , $$ and then set this value equal to $0.5$: $$ 1-e^{-\frac{r^2}{4}}=1/2\Rightarrow e^{-r^2/4}=1/2\Rightarrow e^{-r^2/4}=e^{\ln(1/2)}\Rightarrow r=2\sqrt{\ln 2}\approx 1.66511...\ , $$ as you've done. So your solution manual must have a typo.

  • $\begingroup$ Could you give a reference to some background about using the dirac delta function to arrive at your PDF? I've never seen this usage. $\endgroup$ – lstbl Feb 28 '16 at 14:51
  • 1
    $\begingroup$ @lstbl you can start here pvamu.edu/Include/Math/AAM/Vol3_No1/… and www.ncc.org.in/download.php?f=NCC2008/2008_A5_1.pdf ... once you go Dirac, you never go back! $\endgroup$ – Pierpaolo Vivo Feb 28 '16 at 15:20

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.