I want to create a function f(x) defined on the interval [0..1], which has all of the following properties on the interval:

  1. It is continuous and differentiable throughout the interval, and can be expressed in a closed form without integrals, or through the use of piecewise functions.
  2. $f'(x)$ is never negative for any value of $x \in [0..1]$
  3. $f(0)$ is 0 and $f(1)$ is 1. and $f''(x)$ cannot change sign more than once in the interval.
  4. If one were to select an $x$ randomly from a perfectly uniform distribution on $[0..1]$, the limit on the average $f(x)$ over an infinite number of tries approaches some particular value $c$.

Given some real value for $c \in (0..1)$, what is the simplest way to generate an $f(x)$ that satisfies the criteria?

The first three criteria are fairly easy to achieve... consider the rational function: $$g(x) = \frac{xk}{2kx-k-x+1}$$ satisfies the first 3 properties, guaranteeing intersection with the point $\left(\frac{1}{2},k \right)$, and even satisfies the 4th property for special case where $k=c=\frac{1}{2}$.

I do not know, however, how to satisfy the 4th property in general. I am thinking that it may simply be some kind of modification to the above function, but don't know for sure.

Do I need to find a value of $k$ for the above function $g(x)$ such the closed integral of $g(x)$ for $x=0..1$ = $c$? This might be possible for me to calculate, I haven't tried yet, but if it doesn't yield a function that satisfies the fourth property, I fear it would ultimately only be a waste of time. I am hoping there may a more elementary solution.


Literally just $x^n$ LOL.

For nonnegative $n$, as $n$ increases, the average value is closer to zero, and as $n$ decreases towards zero, the average value is closer to $1$.

All other three criteria are fulfilled too.

  • $\begingroup$ Pretty sure you can’t really get c=0 or c=1, but they are APPROACHABLE $\endgroup$ – Saketh Malyala Jul 1 '17 at 2:27
  • $\begingroup$ But it might not be differentiable at x=0 hm $\endgroup$ – Saketh Malyala Jul 1 '17 at 2:28
  • $\begingroup$ How would you go about calculating 'n' for a given 'c'? $\endgroup$ – Mark Jul 1 '17 at 4:54
  • $\begingroup$ gah... stupid question. I figured it out... n=(1-c)/c. Indeed, that does seem to produce to correct result. $\endgroup$ – Mark Jul 1 '17 at 5:04
  • $\begingroup$ hey that isn't a stupid question! plus you figured it out. pat on the back $\endgroup$ – Saketh Malyala Jul 1 '17 at 5:08

Yes, you need $c$ to be the integral of the function over the interval. For your function, the integral is $$\frac{k \left(2 k-2 (k-1) \tanh ^{-1}(1-2 k)-1\right)}{(1-2 k)^2},$$ so solving for $k$ is not trivial.


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.