Let $X_1, X_2, X_3, X_4$ be independent and $\operatorname{Exp}( \lambda )$. We have, $$Y_1 := X_1+X_2+ X_3+ X_4$$ $$Y_2:= \frac{X_1 }{X_1+X_2}$$ $$Y_3:= \frac{X_1 + X_2 }{X_1+X_2+ X_3}$$ $$Y_4:=\frac{X_1 + X_2 + X_3}{X_1+X_2+ X_3+ X_4}$$ We are now supposed to show that these, $Y_i$ are independent and determine their joint distribution (of $(Y_1, Y_2, Y_3, Y_4)$).

My initial approach is to use the fact that, $$Y_i = g_i(X_1,\dots,X_n)$$ $$X_i = h_i(Y_1,\dots,Y_n)$$ $$f_\mathbf{Y}(y_1,\dots,y_n) = f_\mathbf{X}(h_1(y_1,\dots,y_n),\dots,h_n(y_1,\dots,y_n)) \cdot |J|$$ $J$ being the Jacobian determinant. I start of with re-writing each $X_i$ as a function of $(Y_1,\dots, Y_1)$ and get, after some manipulations that, $$X_1 = Y_1 Y_2 Y_3 Y_4$$ $$X_2 = Y_1 Y_3 Y_4 (1-Y_2)$$ $$X_3 = Y_1 Y_4(1-Y_3)$$ $$X_4 = Y_1(1-Y_4)$$

Computing all partial derivatives,$\frac{\partial x_j}{\partial y_i}$ , yields me the following matrix,

\begin{bmatrix} y_2y_3y_4 & y_1y_3y_4 & y_1y_2y_4 & y_1y_2y_3 \\ y_3y_4(1-y_2) & -y_1y_3y_4 & y_1y_4(1-y_2) & y_1y_3(1-y_2) \\ y_4(1-y_3) & 0 & -y_1y_4 & y_1(1-y_3) \\ 1-y_4 & 0 & 0 & -y_1 \end{bmatrix}

My linear algebra is a little rusty, but I believe the doing the following is valid. The aim is to perform row/column operations to obtain a triangular determinant. First I swap the columns: 1. Swap column #1 and #4. 2. Swap column #2 and #1, 3. Swap column #2 and #3. This yields

\begin{bmatrix} y_1y_3y_4 & y_1y_2y_4 & y_1y_2y_3 & y_2y_3y_4 \\ -y_1y_3y_4 & y_1y_4(1-y_2) & y_1y_3(1-y_2)& y_3y_4(1-y_2) \\ 0 & -y_1y_4 & y_1(1-y_3) & y_4(1-y_3) \\ 0 & 0 & -y_1 & 1-y_4 \end{bmatrix}

I then: 1. Add row #1 to row #2. 2. Add row #2 to row #3. 3. Add row #3 to row #4. I then have,

\begin{bmatrix} y_1y_3y_4 & y_1y_2y_4 & y_1y_2y_3 & y_2y_3y_4\\ 0 & y_1y_4 & y_1y_3 & y_3y_4 \\ 0 & 0 & y_1 & y_4 \\ 0 & 0 & 0 & 1 \end{bmatrix} If I've done this correctly then, $$ |J| = y_1^3 y_4^2 y_3 $$ We have that the joint p.d.f of $(X_1,X_2,X_3,X_4)$ is, $$ f_\mathbf{X} = \lambda^4 e^{-\lambda(x_1+x_2+x_3+x_4)} $$ We then get, after some manipulation $(x_1 + x_2 + x_3 + x_4 = y_1)$, that, $$ f_\mathbf{Y} = \lambda^4 e^{-\lambda y_1} y_1^3 y_4^2 y_3 $$

The question is now, have I done this correctly (i.e. computed the joint distribution of $(Y_1, Y_2, Y_3, Y_4)$? If so, how do I show independence for the $Y_i$'s? Do I have to compute all the marginal distributions or is there a simpler approach?

Any hints on how to proceed would be greatly appreciated!


  • 1
    $\begingroup$ looks ok (to me), you might write supports of $y_1,\dots,y_4$. $\endgroup$
    – MAN-MADE
    Sep 17 '17 at 16:29
  • $\begingroup$ I think it's OK, too. You might offer, as a check, that the integral of your joint density over the range in question works out to 1. $\endgroup$ Sep 17 '17 at 16:31
  • $\begingroup$ So you have $y_1\ge0,\ 0\le y_2\le 1,\ 0 \le y_3\le 1,\ 0\le y_4\le 1.$ I wouldn't omit that from any write-up explaining the solution. $\qquad$ $\endgroup$ Sep 17 '17 at 16:33
  • $\begingroup$ Thank you!! I integrated all $y_i$ over $[ 0; \inf [$ and couldn't get the product to equal one. This solved it! $\endgroup$
    – gabruomad
    Sep 17 '17 at 22:48

If what you've done is correct (I haven't checked the details), then independence of $Y_1,Y_2,Y_3,Y_4$ follows from the factorization of their joint density: It's the product of a function of $y_1,$ a function of $y_2,$ a function of $y_3,$ and a function of $y_4.$ That's enough to infer independence. (If it depended on something like $y_2^2+y_2^2$ or $\cos(y_1+y_2),$ etc., then it would not be such a product. The function of $y_2$ in this case is constant, and from that you conclude that $Y_2$ is uniformly distributed.


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.