Representation for increasing function on unit square

330 Views Asked by At

Do you know any result concerning the representation of functions $f : [0,1] \to [0,1]$ continuous and increasing (with $f(0)=0,f(1)=1)$) as convex combinations of a family of particular functions?

If we denote by $\mathcal{K}$ the family of functions like above the question is if there exists a family $\mathcal{B}$ (strictly smaller than $\mathcal{K}$) of functions as above such that each $f \in \mathcal{K}$ can be written as $$ \sum_{i \in I}\alpha_i b_i \text{ with } b_i \in \mathcal{B} \text{ and } \sum_{i\in I}\alpha_i=1 (\alpha_i>0) $$


I think a related question is the following:

If we denote by $M=\{f:[0,1] \to [0,1]: f(x)= x^\lambda , \lambda > 0\}$ then what is the convex hull of $M$? Is it close to the family $\mathcal{K}$ above? I guess it cannot be equal, since curves which simultaneously have low slope around zero and low slopes around one are not in $M$.


If we view $\mathcal{K}$ as a subset of $C([0,1])$ we can easily see that it is convex. Is there any way we can impose some restrictions on $\mathcal{K}$ ($C^1$ with bounded derivatives, Lipschitz, etc) so that the resulting set is still convex and has reasonable extremal points (in order to use Krein Milman theorem)?

1

There are 1 best solutions below

4
On

The subset of piecewise linear functions is such a subset. Take an increasing piece wise linear function $f$ that fixes 0 and 1 such that $1/2 f$ stays within 1/2 of the function while remaining below it (such as being 0 until the function is close to 1/2, then rising to 1/2 until the end). Then take another piece wise linear function $g$ such that $1/4 g+1/2 f$ stays within 1/4 of the function, while still remaining below it. Continuing in this way, we get the original function as a convex combination of piece wise linear functions.

A certain subset of increasing piece wise linear functions fixing the end points is called Thompson's group.

Edit As pointed out in the comments, this doesn't work. This construction works for $C^1$ functions, and possibly Lipschitz or absolutely continuous functions.