I would like to have a function: $$ f: \mathbb{R} \rightarrow \left\{x \in \mathbb{R}^n \mid \sum_{i=1}^n x_i = 1\right\}:x \mapsto (f_1(x),\dots,f_n(x)) $$ s.t. $f(0) = (1,0,\dots,0)$ and $f(1) = (0,\dots,0,1)$. This is of course very easy to find but I would like to have these such that the conversion happens in a smooth way. What I mean by that is that I would like to have the following extra properties:

  • $f_1(1/2) = \dots = f_n(1/2) = 1/n$
  • $f_1(x)$ is descending
  • $f_n(x)$ is ascending
  • If we define the function $g_x(k) := f_k(x)$ and we draw the graphs of $g_x(k)$ for $x$ going from $0$ to $1$ this should look like a bit like a wave (idea further explained in next two points)
  • On $]0,1/2[$ we have $f_1(x) > \dots > f_n(x)$ and on $]1/2,1[$ we have $f_1(x) < \dots < f_n(x)$

Example method

For example for $n = 2$ we can define: $$ f(x) := (x, 1-x) $$ now for $n = 3$ we can write: $$ f(x) := (x,f_2(x),1-x) $$ but then we have a problem as we need to have $f_2(x) = 0$, therefore we need to let $f_1(x)$ descend to $1/3$ on $[0,1/2]$ and $f_3(x)$ ascend to $1/3$ on $[0,1/2]$ we define: $$ f_1(x) :=\begin{cases} (1 - \frac{4}{3} \cdot x) & \mbox{ if } x \in [0,1/2]\\ \frac{2}{3}(1 - x) & \mbox{ if } x \in [1/2,1] \end{cases} $$ and $$ f_3(x) :=\begin{cases} \frac{2}{3} \cdot x & \mbox{ if } x \in [0,1/2]\\ \frac{4}{3} x - \frac{1}{3} & \mbox{ if } x \in [1/2,1] \end{cases} $$ then we automatically find for $f_2(x)$: $$ f_2(x) = \begin{cases} \frac{2}{3} x & \mbox{ if } x \in [0,1/2]\\ \frac{2}{3}(1 - x) & \mbox{ if } x \in [1/2,1] \end{cases} $$

Now we can continue for $n = 4$, the logical way to continue would be to define $f_1(x)$ to be $(1 - \frac{3}{2} x)$ on $[0,1/2]$ and $f_4(x) := \frac{1}{2} x$ on $[0,\frac{1}{2}]$. We could then use $f_2 = f_3$ and solve the equation $f_1 + 2 f_2 + f_4 = 1$ on $[0,\frac{1}{2}]$. But this does not satisfy the last property and doesn't look enough like a wave to me.

Reason For My Question

The Reason I ask this is because I would like to have a sequence of Markov Chains which goes from perfect correlation (i.e. transition matrix the unity matrix) to some other Markov Chain which is strongly negatively correlated.

  • $\begingroup$ Do you need $\sum f_i(x)=1$ always, or just at $x=0,{1\over2},1$? $\endgroup$ Jan 19, 2017 at 19:24
  • $\begingroup$ Always, I'll add the reason for my qustion which might make things more clear. $\endgroup$
    – HolyMonk
    Jan 19, 2017 at 19:25
  • $\begingroup$ Some comments: I think you want $f_1$ descending (decreasing) and $f_n$ ascending (increasing). Also, due to the symmetry of the problem, you can simplify some of the requirements, by looking for a function $f:[0,1/2]\to\mathbb{R}^n$ satisfying the required properties on the half-interval, then for $t\in(1/2,1)$ put $f_k(t)=f_{n+1-k}(1-t)$ for $k=1,\ldots,n$. $\endgroup$
    – Aweygan
    Jan 19, 2017 at 19:30
  • $\begingroup$ I fully agree, made the changes to the question. $\endgroup$
    – HolyMonk
    Jan 19, 2017 at 19:31

1 Answer 1


$$f_k(x)={q^{k-1}\over\sum\limits_{i=0}^{n-1}q^i},\text{ where }q={x\over1-x}$$ would do the job.

  • $\begingroup$ Perfect, I plotted it and it's the wave I was looking for, thanks! $\endgroup$
    – HolyMonk
    Jan 19, 2017 at 19:55

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.