I want to solve this exercise with the transformation formula, what did I do wrong in my solution?:

Let $X,Y,Z$ be independent random variables with uniform distribution on [0,1]. What's the expectation value of $(X+Y+Z)^2$


The transformation formula is $$f_{W,Y,Z}=f_{X,Y,Z}|\operatorname{det}(J_{T^{-1}})|$$ to calculate a joined density. Lets define a transformation $T$

$T(X,Y,Z)=\left((X+Y+Z)^2,Y,Z\right)\qquad T([0,1],[0,1],[0,1])={[0,9],[0,1],[0,1]}$

Then $T^{-1}(W,Q,R)=(\sqrt{W}-Q-R,Q,Z)$

The jacobi matrix is

$$J_T^{-1} = \begin{pmatrix}\frac{1}{2\sqrt{w}} & -1 & -1\\ 0 & 1 &0\\ 0 & 0 & 1\end{pmatrix}$$

$$det( J_{T^{-1}}) = \frac{1}{2\sqrt{w}} $$ $$f_{W,Q,R}=f_{X,Y,Z}(\sqrt{W}-Q-R,Q,R)\frac{1}{2\sqrt{w}}=\mathrm{1}_{q\in(0,1)}\mathrm{1}_{r\in(0,1)}\frac{1}{2\sqrt{w}}=\mathrm{1}_{r\cap q\in(0,1)}\frac{1}{2\sqrt{w}}$$

So we integrate with respect to $q$ and $r$ $$f_{W}=\int\limits_{0}^1 \int\limits_{0}^1\frac{1}{2\sqrt{w}}\mathrm{d}r\mathrm{d}q$$

Now the expectation value is $$E[W]=\int\limits_{0}^9 w f_{W}\mathrm{d}w= \int\limits_0^9 \frac{w}{2 \sqrt{(w)}} \mathrm{d}w = 9$$

The solution says it's $5/2$ and I know that I can solve it with the expansion of $(X+Y+Z)^2$, but I want to solve it with the transformation rule so what did I do wrong?

  • 2
    $\begingroup$ Your $f_{WQR}$ is incorrect - it needs a third indicator term $\mathbf{1}_{\{0<\sqrt{w} - q -r <1\}}$. Also, this is a needlessly complex way to go about this. Exploit the independence of the RVs to directly calculate the second moment, the joint density is not required. <Include rant about appropriate tool for the job> $\endgroup$ – stochasticboy321 Jul 11 '16 at 11:48
  • $\begingroup$ @stochasticboy321 Thank you very much for the answer. But I don't understand why I need this indicator function that depends on all 3 variables, can I not just put the interval $[0,1]^3$ into the transformation and then I get $ [0,9]\times [0,1]\times[0,1]$? $\endgroup$ – Matriz Jul 11 '16 at 15:33

I think you are quite over-complicating it.
If $X,Y,Z$ are uniformly distributed over $[0,1]$ and independent, $$\begin{eqnarray*}E=\mathbb{E}[(X+Y+Z)^2]&=&\iiint_{[0,1]^3}(x+y+z)^2\,d\mu\\&=&\iiint_{[0,1]^3}(x^2+y^2+z^2+2xy+2xz+2yz)\,d\mu\\&=&3\cdot\frac{1}{3}+3\cdot 2\cdot \frac{1}{4}=\color{red}{\frac{5}{2}}.\end{eqnarray*}$$ Jensen's inequality gives $E\geq \mathbb{E}[X+Y+Z]^2 = \frac{9}{4}$, and the linearity of expectation gives $$ E = 3\mathbb{E}[X^2]+6\mathbb{E}[X]^2 =3\text{Var}[X]+9\mathbb{E}[X]^2$$ so $E-\frac{9}{4}$ is exactly $3\text{Var}[X]=\frac{1}{4}$.

  • 2
    $\begingroup$ @Matriz Yes. If you know the joint density function, $f_{X,Y,Z}(x,y,z)$, and the support interval, $\mathcal S$, then indeed, $\mathbb E[g(X,Y,Z)] = \iiint_{\mathcal S} g(x,y,z)\,f_{X,Y,Z}(x,y,z)\operatorname d z \operatorname d y \operatorname d x$ $\endgroup$ – Graham Kemp Jul 11 '16 at 13:08
  • $\begingroup$ @GrahamKemp Thanks a lot! $\endgroup$ – Matriz Jul 11 '16 at 14:51

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.