James from the MacMillan Academy was the
first to crack this problem - well done James!
Steve says
The essence of this problem lies in the fact that a cdf $F(x)$ is
non-decreasing and satisfies $0\leq F(x) \leq 1$, whereas a pdf is
a non-negative function which integrates to 1 between $-\infty$ and
$\infty$. The areas under the curves are the key points to
consider. A cdf $F(x)$ can either reach the value $1$ for a finite
value of $x_1$ or tend to $1$ as $x\to \infty$. In the first case,
provided that the area under the cdf to the left of $x_1$ is 1 then
this will work as a pdf for a random variable which is $0$ if
$x> x_1$. In the second case, the area diverges and, therefore,
cannot be used as a pdf.
To get the curves to match exactly for values of $x< x_1$ we
will need to solve the equation $F'(x) = f(x)$.
Here is James's solution
Part 1
A CDF is always an increasing function which goes to $0$ as $X$
goes to $-\infty$ and 1 as x goes to $\infty$ and a pdf is a
function that is never negative and has an area under it of
1.
The CDF can reach 1 at any point; let us only consider ones that
reach it at 0 as all others either can be generalised from these by
moving them left or right i.e. considering $F(x-a)$ where $a$ is
the point where it reaches one, or they don't reach 1 until
infinity which I shall deal with later.
As the PDF can do what it likes when $x$ is positive (so long as it
stays non-negative and doesn't let the area exceed 1) it can make
up difference between the area to the left of x=0 and 1 and so the
area to the left of x=0 need only be between 0 and 1. So the only rules controlling the CDF
are that to the left of where it reaches F(x)=1 the area underneath
it must be less than or equal to 1 which is true for any of a
multitude of CDF's
Regarding the ones that don't reach 1 until $\infty$ these would
have to be a solution to PDF=CDF for $x< \infty$, consider any
finite value of $x$ if $F(x)> 0$. At that point then it must be
that value or greater until $x=\infty$ giving area under it $\geq
F(x)\times \infty$ which is absurd and definitely larger than 1
unless $F(x)$ is infinitesimal for all $x$ in which case $F(x)$
wouldn't be tending to 1 and so such a CDF cannot possibly be a
PDF.
Part 2
The integral of a CDF from $-\infty$ to $\infty$ is always
$\infty$. Thus a CDF cannot be used as a pdf.
Part 2
As $f(x)=F'(x)$ for $f(x)=F(x)$ we would need $F(x)=F'(x)$. Thus,
writing $F(x)=y$, the CDF must be a solution to the differential
equation