I am still fiddling with the understanding and application of the fractional integration/differentation.
I've tried the wikipedia-formula for the Cauchy's iterated integration:
$$ (J^{\alpha} f)(x) = \frac 1{\Gamma(\alpha)} \cdot \int_0^x((x-t)^{\alpha-1}f(t)) dt$$ and inserted
$$ (J^{1/2} f)(x) = \frac 1{\Gamma(1/2)} \cdot \int_0^x (x-t)^{-1/2} f(t) dt$$
applied to the function $f(x) = \exp(x)-1$ .
Two time the semi-integral for some interval $0 \ldots x$ should reproduce the usual integral, but I get a difference for any $x$ that I try.
Q: Where is the error in the understanding/implementation of the half-integral?
This is, what I've done in Pari/GP:
semiInt(x,d=1e-64)= 1/gamma(1/2) * intnum(t=0,x-d,(x-t)^(1/2-1)*(exp(t)-1))
\\ introduce a delta-deviation "d" for the upper bound to prevent
\\ evaluation with 0^(-1/2) when the formal parameter t=x
x1=0.5 \\ = 0.500000000000
x2 = semiInt(semiInt(x1)) \\ = 0.161464182768
x3 = intnum(t=0,x1,exp(t)-1) \\ = 0.148721270700
x2-x3 \\ = 0.0127429120679
Symbolically... Let $f(x) = e^x-1$. Then I get $$ J^1f(x) = e^x-x-1, \\ f_2(x):= J^{1/2}f(x) = \frac{e^x \sqrt{\pi}\;\text{erf}\sqrt{x} - 2\sqrt{x}}{\sqrt{\pi}}\\ J^{1/2}f_2(x) = \frac{1}{\pi}\int_0^x\frac{e^t \sqrt{\pi}\;\text{erf}\sqrt{t} - 2\sqrt{t}}{\sqrt{x-t}}\;dt . $$ If we did not know where this integral came from, perhaps we would be amazed by the assertion $$ \frac{1}{\pi}\int_0^x\frac{e^t \sqrt{\pi}\;\text{erf}\sqrt{t} - 2\sqrt{t}}{\sqrt{x-t}}\;dt = e^x-x-1 $$ But this is simply $J^1 f = J^{1/2} J^{1/2} f$, which is how the fractional integral is supposed to act.