I've been banging my head against the wall on this one.
Suppose I have 3 numbers $d_1$, $d_2$, and $d_3$. For example's sake, let's say $(d_1, d_2, d_3) = (2, 3, 4)$ (if I can solve this, then I think I'd know how to generalize). I want to create 3 random variables $r_1, r_2,$ and $r_3$ such that
- $r_1 - r_2$ is identically the uniform distribution $U(-d_1, d_1)$,
- $r_2 - r_3$ is identically the uniform distribution $U(-d_2, d_2)$, and
- $r_3 - r_1$ is identically the uniform distribution $U(-d_3, d_3)$.
If you look at my history you can see a way to do it if say $d_1 = d_2 = d_3$. Solving other cases with $d_i\in\{1, 2, 3\}$ is all readily achievable, but even with a silly case like $(d_1, d_2, d_3) = (2, 3, 4)$ I'm quite stuck.
I'd appreciate any thoughts.
Wew this one took awhile. But here's a way to do it with $d_1 = 2, d_2 = 3$, and $d_3 = 4$:
$$p_3(x) = \begin{cases} 3x+6 & \text{if }x<-1\\ -3x & \text{if }-1<x<1\\ 3x-6 & \text{if }1<x \end{cases}$$ Otherwise, if $c$ is Tails, set
$$p_3(x) = \begin{cases} -3x-6 & \text{if }x<-1\\ -3x & \text{if }-1<x<1\\ -3x+6 & \text{if }1<x \end{cases}$$
One can verify with conventional methods that all pairwise distributions satisfy the above requirements.
I've included below a graphical view of the above setting. The dashed red lines correspond to the $\frac{1}{2}$ probability of $p_3$ following either one.