# Why this limit of integration?

I've been solving problems from the book by DeGroot and Schervish and I can't understand why m is the upper limit of integration in the solution to this problem. Why not the lower one?

Here is the problem:

Suppose that a random variable X has a continuous distribution for which the p.d.f. f is as follows:

$f(x) = 2x$ for $0< x <1, 0$ otherwise

Determine the value of $d$ that minimizes $E(|X − d|)$.

Here is the solution:

$$\int_0^m 2x \, dx=0.5$$

Thank you very much in advance.

-
The answer from nbubis doesn't explain the last integral you gave. I've posted an answer that does. Setting that integral to $0.5$ is not "the solution", but it leads you quickly to the solution. – Michael Hardy Aug 18 '12 at 16:43
I find much confusion in the question and the answers. – Michael Hardy Aug 18 '12 at 21:30
...and now there's somewhat less confusion. – Michael Hardy Aug 18 '12 at 21:51
ahaha;) cool! mathematics doesnt like confusion although it has this inside of his blood.. – Seyhmus Güngören Aug 18 '12 at 21:54

The expected value is given by (assuming $d$ is in $[0,1]$): $$E = \int_0^1{2x|x-d|}dx = \int_0^d{2x(d-x)}dx + \int_d^1{2x(x-d)}dx$$ $$= -\frac{2}{3} + d - \frac{2d^3}{3}$$ Derive by $d$ and compare to zero: $$E' = 1 - 2d^2 = 0 \rightarrow d = \frac{1}{\sqrt{2}}$$

-
Thank you very much for the thorough explanation, nbubis! – Alan Aug 18 '12 at 10:56
I got $2d^2-1$, and so should you. – Michael Hardy Aug 18 '12 at 21:29
@MichaelHardy yes you are right. WHILE I posted my answer I did it based on nbubis's answer. I didnt check it by myself. I only focused on the second part of the question which say minimum! although it seemed to be maximum based on his answer. But you are completely right. – Seyhmus Güngören Aug 18 '12 at 21:41

The value of $d$ that that minimizes $E(|X-d|)$ is the median value of $X$, i.e. the value $m$ such that $$\Pr(X\le m) = 0.5 = \Pr(X\ge m).$$

Thus for continuous distirbutions it is the value of $m$ for which $$\int_{-\infty}^m f_X(x)\,dx = 0.5 = \int_m^\infty f_X(x)\,dx.$$

In your case, you want $$\int_0^m 2x\,dx = 0.5 = \int_m^1 2x\,dx.$$ So $m^2 -0^2 = 0.5 = 1^2-m^2$. Consequently $m= \sqrt{0.5}= \sqrt{2}/2$.

-

I will note one important thing about the question. You will get $$E^{'}[X]=1-2d^2$$ as @nbubis wrote down already. When you find $$E^{''}[X]=-4d$$ and since $d\in[0,1]$ you have $$E^{''}[X]<0$$ which indicates that you have a maximum instead of minimum as you stated in the question EDIT: Result was based on $$E^{'}[X]=1-2d^2$$ Since it is $$E^{'}[X]=2d^2-1,$$ there is a minimum. I didnt solve the first derivative by myself just copied.

-
Thank you very much for clarification! – Alan Aug 18 '12 at 11:40
@Alan my pleasure;) – Seyhmus Güngören Aug 18 '12 at 11:41
This is nonsense. You get a minimum, not a maximum. The derivative should be $2d^2-1$, not $1-2d^2$. – Michael Hardy Aug 18 '12 at 21:30