# Statistics expectation problem involving circle.

1. May 13, 2013

### peripatein

Hi,
1. The problem statement, all variables and given/known data
A circle of radius r is as shown in the attached diagram. I am asked to first express X as a function of θ, then to compute E(X). It is also stated that θ obeys U[0,2π].

2. Relevant equations

3. The attempt at a solution
Through simple trigonometry I have found X = 2rsin(θ/2). As θ distributes evenly, I know that the density of θ should be 1/2π. What is the density of x then? Am I supposed to simply substitute 2arcsin(x/2r)? Or should it be approached by first computing E(θ) and trying to hence infer E(X)?
I'd appreciate an advice.

#### Attached Files:

• ###### Circle.jpg
File size:
24.5 KB
Views:
52
2. May 13, 2013

### tiny-tim

hi peripatein!

look at the definition

if X(θ) is a function of θ, and if the distribution of θ is f(θ),

then what is the definition of E(X) ?

3. May 13, 2013

### peripatein

The definition of E(X) is integral [would the boundaries be 1 to 2r?] x*f(x) dx, where f(x) is the probability density of X. Is this correct?

4. May 13, 2013

### tiny-tim

yees, but it's f(θ) that you know, not f(x) …

so what is E(X) expressed as an integral over θ ?

5. May 13, 2013

### peripatein

E(X) = ∫(2rsin(θ/2)/2π)(rcos(θ/2))dθ from 0 to 2π?

6. May 13, 2013

### tiny-tim

no

stick to the definition for the moment …

what is the definition of E(X) if X = X(θ), and the distribution of θ is f(θ) ?

7. May 13, 2013

### peripatein

I am not sure, tiny-tim :-(.
Should I look for the inverse function of X(θ)?
I mean, the definition reads E[g(X)] = integral (from -inf. to +inf.) g(x)f(x)dx. I understand that as E[θ(X)] = θ(X)*f(x)dx. I am probably mistaken, yet I am not sure how to approach it.
Here's another attempt, which makes not much sense to me but for the sake of trying: might it be E(X) = int X(θ)*f(θ)dθ?

8. May 13, 2013

### tiny-tim

yes, that's correct …

E(X) = ∫ X(θ) f(θ) dθ

… this is exactly the same as the definition you mentioned earlier, ∫ x f(x) dx, but using the distribution of θ rather than the distribution of x (remember, the two f's are different!)

so in this case, E(X) = ∫ 2rsin(θ/2) f(θ) dθ,

which is … ?

9. May 13, 2013

### peripatein

4r/pi?
How may I now determine whether X obeys an even distribution U?

10. May 13, 2013

### tiny-tim

show how you got it

11. May 13, 2013

### peripatein

Well, shouldn't it be E(X) = ∫ 2rsin(θ/2) f(θ) dθ = ∫ (r/pi)sin(θ/2) dθ between 0 and 2pi?
How may I now determine whether X obeys an even distribution U?

12. May 13, 2013

### tiny-tim

correct
it obviously doesn't

θ is evenly distributed, so sinθ/2 isn't

(why are you asking? doesn't the question only ask for E(X) ? )

13. May 13, 2013

### peripatein

Now, the second part of that question, which is not related to the first, asks for a proof that
∫ (from 0 to ∞) xne-x dx = n! for every n > 0, and the hint is to find a recursive argument for In and an initial condition.
I have found In = (In + 1 + xne-x + nxn-1e-x) / (n(n-1)), where x changes from 0 to ∞, but am not sure how to proceed. Could you advise?

14. May 13, 2013

### tiny-tim

(how did you get that ?? )

anyway, you shouldn't have any x's in the equation

x is a dummy variable, the only "real" variable is n​

15. May 13, 2013

### peripatein

Well, integration by parts, twice. Setting the original integral to be In, I let u = xn and dv be e-x. In the second integration by parts I let u = xn-1 and dv = e-x.
Does that make more sense now? Is that how it is to be handled? The boundaries should be between 0 and infinity, but I don't know how to incorporate that into the expression (I mean, what do you get for (xne-x + nxn-1e-x) between 0 and inf.? How do you evaluate that?

16. May 13, 2013

### tiny-tim

why not just once?
i] what is xne-x when x = 0 ?

ii] what is (the limit of) xne-x when x = ∞ ?

17. May 13, 2013

### peripatein

You're right, silly me! A single integration by parts led me to the right answer: In+1 = (n+1)In.
Now, suppose I define f(x) = (xne-x)/n!. I have found E(X) = n + 1 and V(X) = 0.
Would you please confirm that?

18. May 13, 2013

### tiny-tim

yes that looks correct
what is V(X) ?

19. May 13, 2013

### peripatein

Variance (=E(x^2) - (E(x))^2)
Is it not zero in this case?

20. May 13, 2013

### tiny-tim

no

how did you get E(x2) ?