The Riemann-Liouville fractional derivative of order $1/2$ of $x^{-1}$ is defined as
$$
\mathcal{D}_x^{1/2} x^{-1} = \frac{1}{\Gamma(1/2)} \frac{d\;}{dx} \left[\int_0^x (x-\tau)^{-1/2}\tau^{-1} \,d\tau\right] .
$$
Since the above integral does not converge, I understand that this fractional derivative does not exists. However, when I ask Mathematica to calculate it using the command FractionalD, I obtain that the result is
$$
\frac{-\log{x}+ \gamma+\psi^{(0)}(-1/2)}{2\sqrt{\pi} x^{3/2}}
$$
where $\gamma$ is the Euler-Mascheroni constant and $\psi^{(0)}$ is the digamma function.
How did Mathematica arrive to this result?
$\begingroup$
$\endgroup$
9
1 Answer
$\begingroup$
$\endgroup$
7
This is rather a math than Mathematica question. As I understand it, the formula
FractionalD[f[x], {x, a}]==FractionalD[Integrate[f[x], x], {x, a + 1}]
which holds onder certain assumptions is used to this end i.e. to define the fractional derivative of order $1/2$ of $1/x$ (I don't have a reference at hand.). Indeed,
FractionalD[Integrate[1/x, x], {x, 3/2}
(EulerGamma - Log[x] + PolyGamma[0, -(1/2)])/(2 Sqrt[\[Pi]] x^(3/2))
where the integral converges. Hope that gap in the documentation will be filled in next versions of Mathematica.
-
$\begingroup$ The formula
FractionalD[f[x], {x, a}]==FractionalD[Integrate[f[x], x], {x, a + 1}]
can be proven by integration by parts. $\endgroup$ Commented Aug 29 at 12:14 -
$\begingroup$ In the current documentation reference.wolfram.com/legacy/language/v14/ref/FractionalD.html under "Scope" is an example
FractionalD[x^n, {x, 1/3}]
which also includes the case n=-1. $\endgroup$ Commented Aug 29 at 13:01 -
$\begingroup$ @VaclavKotesovec: Did you think twice before having posted your comment? The question is not about that formula, but about its derivation. The definition from the documentation does not work in the case n=1 as shown in the question. If I have nothing to say, I don't say $\endgroup$ Commented Aug 29 at 14:44
-
$\begingroup$ My comment was not about the question, but about your sentence about "gap in the documentation". And it is a case of n=-1 (not n=1). I will not respond to your further comments. $\endgroup$ Commented Aug 29 at 15:26
-
1$\begingroup$ Thanks folks! I'm so used to the Caputo derivative that I didn't realize this works for the Riemann-Liouville definition. $\endgroup$ Commented Aug 29 at 15:50
FractionalD[1/x, {x, 1/2}] /. x -> .5
, we have out 1.04272. On the other hand, we can evaluateNFractionalD[1/x, {x, 1/2}, .5]
, and have out -605.36. It looks like a bug. :) $\endgroup$FractionalD
:GeneralUtilities
PrintDefinitions@FractionalD` (run this twice) Then go into the definition ofiFractionalD
->FractionalDTable
you will see a bunch ofFractionalDTable
definitions. I believeFractionalD
is using one of theseFractionalDTable
definitions whenFractionalD[1/x, {x, 1/2}]
is called here $\endgroup$GeneralUtilities`PrintDefinitions[FractionalD]
the backtick needs to be between GeneralUtilities and PrintDefinitions. Make sure to runGeneralUtilities`PrintDefinitions[FractionalD]
twice because it will come up blank the first time. Look here at Domen's comment for the same format but forFindSequenceFunction
$\endgroup$iFractionalD -> FractionalDTable
in the forum? $\endgroup$