2

Let $g(x) = x^{-\frac{1}{2}}$ and $f(x) = e^{-\sqrt{x} |\omega|}$. I am trying to find an expression for the $M$-th derivative of their product: \begin{align} \frac{d^M}{dx^M} \left[ f(x) g(x) \right] = \sum_{k=0}^{M} \binom{M}{k} f^{(M-k)}(x) g^{(k)}(x) \end{align} where $M$ is an even integer and $\omega \in \mathbb{R}$ is a constant parameter.

The derivative for $g(x)$ is easy:

\begin{align} \frac{d^{k}}{d x^{k}} g(x) = \frac{(2k-1)!!}{2} x^{-\frac{2k+1}{2}} \end{align}

However, for $f(x)$, it is not so easy, and this is where I get stuck.

For the $M$-th derivative of $f(x)$, Mathematica gives me:

M! DifferenceRoot[
   Function[{\[FormalY], \[FormalN]}, {(2 + 6 \[FormalN] + 
          4 \[FormalN]^2) \[FormalY][1 + \[FormalN]] + 
       4 (1 + \[FormalN]) (2 + \[FormalN]) x \[FormalY][
         2 + \[FormalN]] - \[FormalY][\[FormalN]] Abs[w]^2 == 
      0, \[FormalY][0] == 
      E^(-Sqrt[x] Abs[w]), \[FormalY][1] == -((
       E^(-Sqrt[x] Abs[w]) Abs[w])/(2 Sqrt[x]))}]][M]

But I have no idea what any of that means, and their site is not very helpful...

It seems that the solution will involve some kind of recursive application of the product rule, however, I can't seem to find a pattern after I do a handful of layers...


Question:

  1. What is $\frac{d^{n}}{d x^{n}} f(x) = \frac{d^{n}}{d x^{n}} e^{-\sqrt{x} |\omega|}$ when $n$ is an even integer?
The Dude
  • 1,079
  • 9
  • 20

1 Answers1

4

It's sufficient to consider the case $\omega=1$: if we let $$ \color{DarkBlue}{f_n(x):=e^{\sqrt{x}}\sqrt{x}\frac{d^n}{dx^n}\frac{e^{-\sqrt{x}}}{\sqrt{x}}}, $$ then the needed expression is $$ \frac{d^n}{dx^n}\frac{e^{-|\omega|\sqrt{x}}}{\sqrt{x}}=|\omega|^n\frac{e^{-|\omega|\sqrt{x}}}{\sqrt{x}}f_n(\omega^2 x). $$

To compute $f_n(x)$, we consider (for $x>0$ and $|t|<x$) $$ f(x,t):=\sum_{n=0}^\infty f_n(x)\frac{(-t)^n}{n!}=e^{\sqrt{x}}\sqrt{x}\frac{e^{-\sqrt{x-t}}}{\sqrt{x-t}}=\frac{e^{\sqrt{x}-\sqrt{x-t}}}{\sqrt{1-t/x}}. $$

Now we use the formula (for $|z|<1$) $$\frac{(1-\sqrt{1-z})^k}{\sqrt{1-z}}=2^k\sum_{n=k}^\infty\binom{2n-k}{n}\left(\frac{z}{4}\right)^n.$$ To prove it, one may use the known formula $$ \left(\frac{1-\sqrt{1-4z}}{2z}\right)^k=\sum_{n=0}^\infty\binom{2n+k}{n}\frac{k}{2n+k}z^n; $$ here $k/(2n+k)$ is interpreted as $1$ when $n=k=0$ (recipe: multiply this by $z^k$, replace $z$ with $z/4$ and $k$ with $k+1$, and take the derivative).

Returning to the computation, we have $$ f(x,t)=\sum_{k=0}^\infty\frac{1}{k!}\frac{(\sqrt{x}-\sqrt{x-t})^k}{\sqrt{1-t/x}} =\sum_{k=0}^\infty\frac{(2\sqrt{x})^k}{k!}\sum_{n=k}^\infty\binom{2n-k}{n}\left(\frac{t}{4x}\right)^n, $$ which, after (exchanging the summations and) extracting the coefficient of $t^n$, results in $$ f_n(x)=(-1)^n\sum_{k=0}^n\frac{(2n-k)!}{k!(n-k)!}(2\sqrt{x})^{k-2n} \color{DarkBlue}{=(-1)^n\sum_{k=0}^n\frac{(n+k)!}{k!(n-k)!}(2\sqrt{x})^{-n-k}}. $$


Another approach is to use the following representation: $$ \frac{e^{b(s-\sqrt{a^2+s^2})}}{\sqrt{a^2+s^2}}=\int_0^\infty e^{-st}J_0\left(a\sqrt{t^2+2bt}\right)dt, $$ where $J_0(z)=\sum_{n=0}^\infty(-z^2/4)^n/n!^2$ is a Bessel function.

Here we put $s=\sqrt{x}$, $a=\sqrt{y}$ and $b=1$; the LHS is then $f(x,-y)/\sqrt{x}$, and the RHS may be "evaluated" by substituting the power series for $J_0$ directly, and integrating the result termwise, using the binomial formula for $(t+2)^n$. The outcome is the same.

metamorphy
  • 39,111
  • That is amazing. How in the world does one come up with that? – The Dude Feb 04 '21 at 22:38
  • That pattern is exactly what I was trying to find haha hmmm. I am trying to learn something (like an approach or trick or something) from your answer, other than the result of course, but it looks like just a good catch. – The Dude Feb 04 '21 at 22:45
  • In fact it is. More precisely, the "good catch" is the existence of the power series in $z$ above. – metamorphy Feb 04 '21 at 22:47
  • Wow - Absolutely beautiful! I'm stunned. – K.defaoite Feb 05 '21 at 00:03
  • Where does $\sqrt{y}$ come from? – The Dude Feb 06 '21 at 18:04
  • Also, I can just differentiate under the integral sign to find the $n$-th derivative, correct? – The Dude Feb 06 '21 at 18:08
  • @TheDude: I just need a new variable here ($t$ is already in action), so I take $y$ to be. And yes, you can, but I'm in doubt you really want to ;) The power series does the same in a more straightforward way. – metamorphy Feb 06 '21 at 19:09
  • @metamorphy I suppose I am confused because $|\omega|$ seems to have disappeared. Well, for $n$th derivative, I will first expand both sides in the power series. Then, I just need to integrate the $n$th term of the Bessel function power series to 'match terms' on both sides. Correct? – The Dude Feb 06 '21 at 19:12
  • @TheDude: Yes. Both LHS and RHS are considered as power series in $y$, and we're comparing their coefficients. – metamorphy Feb 06 '21 at 19:15
  • Nice solution, for sure. And then ........$(+1) $ for sure. ! Cheers :-) – Claude Leibovici Jun 03 '22 at 08:49