9
$\begingroup$

Let $\lambda$ and $\mu$ be two positive real numbers and let denote $f$ the function defined as:

$$\forall x>0,~f(x):= \exp\left(-\frac{\lambda(x-\mu)^2}{2\mu^2x}\right).$$

I am struggling to find out an explicit expression of $\partial_x^n f$. According to this answer, it seems that:

$$\partial_x^n f(x) = f(x)\frac{P_{\lambda, \mu}(x)}{2^n \mu^{2n}x^{2n}}$$

where $P_{\lambda, \mu}$ is a polynomial function of $x$ depending on $\lambda$ and $\mu$.

Does anyone know how to deal with this derivative ?

Any hint, solution or reference will be highly appreciated!

Thank you.

$\endgroup$
2
  • 1
    $\begingroup$ Of course $P_{\lambda, \mu}(x)$ also depends on $n$. For $n=3$, $$P_{\lambda, \mu}(x) =-8\,{\lambda}^{3}{x}^{6}+24\,{\lambda}^{3}{\mu}^{2}{x}^{4}+96\,{ \lambda}^{2}{\mu}^{4}{x}^{3}-8\,\lambda\, \left( 3\,{\lambda}^{2}{\mu} ^{4}-24\,{\mu}^{6} \right) {x}^{2}-96\,{\lambda}^{2}{\mu}^{6}x+8\,{ \lambda}^{3}{\mu}^{6}$$ $\endgroup$ Commented Apr 21 at 1:04
  • 4
    $\begingroup$ Not sure how to use it but changing the variable to $t=\sqrt{\frac\lambda{2x}}\left(1-\frac x\mu\right)$ should presumably give expressions through Hermite polynomials in $t$ $\endgroup$ Commented Apr 21 at 4:07

8 Answers 8

9
$\begingroup$

$\newcommand\la\lambda$Let $$g(x):=-\frac\la{2\mu^2}\frac{(x-\mu)^2}x =-\frac\la{2\mu^2}(x-2\mu+\mu^2/x)$$ so that $$f(x)=e^{g(x)}.$$ So, by the Faà di Bruno formula, $$f^{(n)}(x)=f(x)\sum_{\pi\in\Pi_n}\prod_{B\in\pi}g^{(|B|)}(x),$$ where $\Pi_n$ is the set of all partitions of the set $\{1,\dots,n\}$, $|B|$ is the cardinality of $B$, and $$g^{(k)}(x)=-\frac\la{2\mu^2}\,1(k=1)-\frac\la2\,(-1)^k k!x^{-k-1}$$ for $k=1,2,\dots$.


For $y_n:=f^{(n)}(x)/n!$, Mathematica gives the recursion $$y(n+2) \left(-\lambda \mu ^2+4 \mu ^2 n x+\lambda x^2+8 \mu ^2 x\right)+2 \mu ^2 (n+3) x^2 y(n+3)+2 y(n+1) \left(\mu ^2+\mu ^2 n+\lambda x\right)+\lambda y(n)=0$$ for $n\ge0$, which implies that $$f^{(n)}(x)=f(x)n!\frac{P_n(x)}{2^n \mu^{2n}x^{2n}},$$ where, for each $n$, $P_n(x)=P_{n;\la,\mu}(x)$ is a polynomial such that $$P_0(x)=1,\quad P_1(x)=\lambda \left(\mu ^2-x^2\right),\quad P_2(x)=\frac{1}{2} \lambda \left(\lambda \mu ^4+\lambda x^4-2 \lambda \mu ^2 x^2-4 \mu ^4 x\right),$$ and the recurrence $$4 \lambda \mu ^4 x^4 P_n(x)+4 \mu ^2 x^2 P_{n+1}(x) \left(\mu ^2 (n+1)+\lambda x\right)+P_{n+2}(x) \left(-\lambda \mu ^2+4 \mu ^2 (n+2) x+\lambda x^2\right)+(n+3) P_{n+3}(x)=0$$ holds for $n\ge0$.

$\endgroup$
1
  • $\begingroup$ Thank you @Iosif Pinelis! $\endgroup$
    – NancyBoy
    Commented Apr 21 at 15:24
6
$\begingroup$

The polynomials $P$ are related to the generalized Laguerre polynomials, see A066667. After some algebra, the $n$-th derivative of $f(x)$ can be written as \begin{align}\label{eq:1}\tag{1} \frac{\partial^n_x f(x)}{f(x)}= \left(-\frac{\lambda }{2\mu ^2}\right)^n + \sum_{k=1}^n \frac{n!}{(n-k)!} \, \frac{(-1)^k \mu ^2 }{k \, x^{k+1}} \left(-\frac{\lambda }{2\mu^2}\right)^{n-k+1} L_{k-1}^{(1)}\left(\frac{\lambda }{2 x}\right), \end{align} with the generalized Laguerre polynomials $L^{(\alpha)}_n(x)$.

$\endgroup$
2
  • 1
    $\begingroup$ Thank you @Fred Hucht for your answer! Could you be a bit more explicit on the main steps of "some algebra" ? Thank you! $\endgroup$
    – NancyBoy
    Commented Apr 22 at 12:13
  • 1
    $\begingroup$ I expanded the LHS of (1) for $n=3,\ldots,7$ using Mathematica's Apart[] (with the substitution $\lambda \mapsto 2 t x$). The resulting terms, e.g., $5040 - 15120 t + 12600 t^2 - 4200 t^3 + 630 t^4 - 42 t^5 + t^6$ for $n=7$, were looked up and identified as generalized Laguerre polynomials using oeis.org. Finally, the prefactors were adjusted. $\endgroup$
    – Fred Hucht
    Commented Apr 22 at 13:00
5
$\begingroup$

$\newcommand\la\lambda$Here is a very simple expression for $f^{(N)}(x)$:

Note that $$f(x)=g(x)h(bx),$$ where $$g(x):=e^{a(x-2\mu)},\quad h(u):=e^{1/u},\quad a:=-\frac\la{2\mu^2},\quad b:=-\frac2\la.$$ So, by the Leibniz differentiation formula, $$f^{(N)}(x)=g(x)\sum_{n=0}^N\binom Nn a^{N-n} b^n h^{(n)}(bx).$$ In turn, it is easily checked by induction on $n$ that $$h^{(n)}(u)=h(u)\sum_{k=0}^n c_{n,k}u^{-k-n}, \tag{10}\label{10}$$ where $$c_{n,k}:=(-1)^n\frac{n!}{k!}\binom{n-1}{k-1}$$ (with $\binom{n-1}{k-1}=0$ for $k=0$).

So, $$f^{(N)}(x)=f(x)\sum_{n=0}^N\binom Nn a^{N-n} b^n \sum_{k=0}^n (-1)^n\frac{n!}{k!}\binom{n-1}{k-1}(bx)^{-k-n}.$$


Note that, in view of the partial fraction decomposition, \eqref{10} allows one to get comparatively simple expressions for the $n$th derivative of any expression of the form $e^{P(x)/Q(x)}$, where $P(x)$ is any complex polynomial and $Q(x)$ is any complex polynomial without multiple roots. Moreover, the no-multiple-roots restriction can be removed, because any complex polynomial with multiple roots and its derivatives can be approximated however closely by a complex polynomial without multiple roots and its derivatives.

Also, writing (say) $\sin\frac1x=\frac1{2i}(e^{-1/(ix)}-e^{1/(ix)})$, we can use \eqref{10} to get a rather simple expression for the $n$th derivative of $\sin\frac1x$.

$\endgroup$
9
  • $\begingroup$ I was starting the same computation! This seems the simplest way to connect the $P'_{\lambda,\mu}$'s with something known $\endgroup$ Commented Apr 23 at 3:26
  • 1
    $\begingroup$ @Iosif Pinelis Nice derivation! Note that for $n>0$, $\sum_{k=0}^n \frac{1}{k!}\binom{n-1}{k-1}(bx)^{-k-n}$ $=$ $\frac{1}{n\,(b x)^{n+1}}L_{n-1}^{(1)}(-\frac{1}{bx})$, with the generalized Laguerre polynomials $L$ from my answer. $\endgroup$
    – Fred Hucht
    Commented Apr 23 at 6:52
  • $\begingroup$ @PietroMajer : Thank you for your comment. $\endgroup$ Commented Apr 24 at 1:26
  • $\begingroup$ @FredHucht : Thank you for your comment. $\endgroup$ Commented Apr 24 at 1:26
  • $\begingroup$ @IosifPinelis: very nice derivation! Thank you $\endgroup$
    – NancyBoy
    Commented Apr 24 at 7:13
4
$\begingroup$

It is easy to see that we can write \begin{equation*} h(x)=\ln f(x)=-\frac{\lambda(x-\mu)^2}{2\mu^2x}=\frac{\lambda}{2\mu^2}\biggl(2\mu-x-\frac{\mu^2}{x}\biggr). \end{equation*} Hence, its derivatives are \begin{equation*} h'(x)=-\frac{\lambda}{2\mu^2}\biggl(1-\frac{\mu^2}{x^2}\biggr) =\frac{\lambda}{2}\biggl(\frac{1}{x^2}-\frac{1}{\mu^2}\biggr) \end{equation*} and \begin{equation*} h^{(k+1)}(x)=(-1)^{k}\frac{\lambda}{2}\frac{(k+1)!}{x^{k+2}},\quad k\in\mathbb{N}. \end{equation*} Making use of the identity \begin{equation*} B_{n,k}(x_1,x_2,\dotsc,x_{n-k+1}) =\frac{n!}{(n-k)!}\sum_{\ell=0}^{k}\frac{x_1^\ell}{\ell!} B_{n-k,k-\ell}\biggl(\frac{x_2}{2}, \frac{x_3}{3},\dotsc,\frac{x_{n+\ell-2k+2}}{n+\ell-2k+2}\biggr) \end{equation*} for $n\ge k\ge0$, where $B_{n,k}$ stands for the partial Bell polynomials (also known as the Bell polynomials of the second kind), we obtain \begin{align*} &\quad B_{n,k}\bigl(h'(x),h''(x),\dotsc,h^{(n-k+1)}(x)\bigr)\\ &=\frac{n!}{(n-k)!}\sum_{\ell=0}^{k}\frac{[h'(x)]^\ell}{\ell!} B_{n-k,k-\ell}\biggl(\frac{h''(x)}{2}, \frac{h'''(x)}{3}, \dotsc, \frac{h^{(n+\ell-2k+2)}(x)}{n+\ell-2k+2}\biggr)\\ &=\frac{n!}{(n-k)!}\sum_{\ell=0}^{k}\frac{1}{\ell!} \biggl[\frac{\lambda}{2}\biggl(\frac{1}{x^2}-\frac{1}{\mu^2}\biggr)\biggr]^\ell B_{n-k,k-\ell}\biggl(-\frac{1}{2}\frac{\lambda}{2}\frac{2!}{x^{3}}, \frac{1}{3} \frac{\lambda}{2}\frac{3!}{x^{4}}, \dotsc, \frac{(-1)^{n+\ell-2k+1}}{n+\ell-2k+2}\frac{\lambda}{2}\frac{(n+\ell-2k+2)!}{x^{n+\ell-2k+3}}\biggr)\\ &=\frac{n!}{(n-k)!}\sum_{\ell=0}^{k}\frac{1}{\ell!} \biggl[\frac{\lambda}{2}\biggl(\frac{1}{x^2}-\frac{1}{\mu^2}\biggr)\biggr]^\ell \biggl(\frac{\lambda}{2}\biggr)^{k-\ell} \frac{(-1)^{n-k}}{x^{n+k-2\ell}} B_{n-k,k-\ell}(1!, 2!, \dotsc,(n+\ell-2k+1)!)\\ &=\frac{(-1)^{n-k}}{(n-k)!}\frac{n!}{x^{n+k}}\biggl(\frac{\lambda}{2}\biggr)^{k} \sum_{\ell=0}^{k}\frac{1}{\ell!}\biggl(1-\frac{x^2}{\mu^2}\biggr)^\ell\binom{n-k-1}{k-\ell-1}\frac{(n-k)!}{(k-\ell)!}\\ &=\frac{(-1)^{n+k}}{x^{n+k}}\frac{n!}{k!}\biggl(\frac{\lambda}{2}\biggr)^{k} \sum_{\ell=0}^{k}\binom{k}{\ell}\binom{n-k-1}{k-\ell-1}\biggl(1-\frac{x^2}{\mu^2}\biggr)^\ell \end{align*} for $n\ge k\ge0$, where we used the identity $$ B_{n,k}\bigl(abx_1,ab^2x_2,\dotsc,ab^{n-k+1}x_{n-k+1}\bigr) =a^kb^n B_{n,k}(x_1,x_2,\dotsc,x_{n-k+1}) $$ and the formula $$ B_{n,k}(1!,2!,\dotsc,(n-k+1)!)=\binom{n-1}{k-1}\frac{n!}{k!} $$ for $n\ge k\ge0$. Further, by the Faa di Bruno formula \begin{equation*} \frac{\operatorname{d}^n[f\circ h(x)]}{\operatorname{d} x^n}=\sum_{k=0}^nf^{(k)}(h(x)) B_{n,k}\bigl(h'(x),h''(x),\dotsc,h^{(n-k+1)}(x)\bigr), \quad n\ge0, \end{equation*} we acquire \begin{align*} f^{(n)}(x)&=\sum_{k=0}^{n}\operatorname{e}^{h(x)} B_{n,k}\bigl(h'(x),h''(x),\dotsc,h^{(n-k+1)}(x)\bigr)\\ &=\operatorname{e}^{h(x)}\sum_{k=0}^{n}\frac{(-1)^{n+k}}{x^{n+k}}\frac{n!}{k!}\biggl(\frac{\lambda}{2}\biggr)^{k} \sum_{\ell=0}^{k}\binom{k}{\ell}\binom{n-k-1}{k-\ell-1}\biggl(1-\frac{x^2}{\mu^2}\biggr)^\ell\\ &=f(x)\frac{n!}{x^{2n}}\sum_{k=0}^{n}\frac{(-1)^{n-k}}{k!}\biggl(\frac{\lambda}{2}\biggr)^{k} x^{n-k} \sum_{\ell=0}^{k}\binom{k}{\ell}\binom{n-k-1}{k-\ell-1}\biggl(1-\frac{x^2}{\mu^2}\biggr)^\ell\\ &=\frac{f(x)}{2^n \mu^{2n} x^{2n}}\sum_{k=0}^{n}\lambda^k (-2x)^{n-k} \sum_{\ell=0}^{k}\binom{n}{\ell} \binom{n-k-1}{k-\ell-1} \frac{(n-\ell)!}{(k-\ell)!} \mu^{2(n-\ell)} \bigl(\mu^2-x^2\bigr)^\ell \end{align*} for $n\ge0$. In conclusion, we obtain the explicit derivative formula $$ \boxed{f^{(n)}(x)=\frac{f(x)}{2^n \mu^{2n} x^{2n}}\sum_{k=0}^{n}\lambda^k (-2x)^{n-k} \sum_{\ell=0}^{k}\binom{n}{\ell} \binom{n-k-1}{k-\ell-1} \frac{(n-\ell)!}{(k-\ell)!} \mu^{2(n-\ell)} \bigl(\mu^2-x^2\bigr)^\ell} $$ for $n\ge0$.

References

  1. C. A. Charalambides, Enumerative Combinatorics, CRC Press Series on Discrete Mathematics and its Applications. Chapman & Hall/CRC, Boca Raton, FL, 2002.
  2. L. Comtet, Advanced Combinatorics: The Art of Finite and Infinite Expansions, Revised and Enlarged Edition, D. Reidel Publishing Co., 1974; available online at https://doi.org/10.1007/978-94-010-2196-8.
  3. https://qifeng618.wordpress.com/2018/03/19/some-papers-on-explicit-formulas-of-special-values-for-the-bell-polynomials-of-the-second-kind/.
$\endgroup$
6
  • 2
    $\begingroup$ Your formula $$f^{(n)}(x)=\frac{e^{-\frac{\lambda (x-\mu)^2}{2 \mu^2 x}}}{2^n \mu^{2 n} x^{2 n}} \sum\limits_{k=1}^n \lambda^k (-2 x)^{n-k} \sum\limits_{\ell=0}^k \binom{n}{\ell} \binom{n-k-1}{k-\ell-1} \frac{(n-\ell)!}{(k-\ell)!} \mu^{2 (n-\ell)} \left(x^2-\mu^2\right)^{\ell}$$ doesnt seem to evaluate correctly for $n>0$ even after I modified the outer sum to start at $k=0$ instead of $k=1$. $\endgroup$ Commented Apr 22 at 0:09
  • $\begingroup$ @StevenClark I have checked up several times, but didn’t find where the errors locate. I also computed by another approach and verified by Mathematica, I haven’t located where errors appear yet. I hope to correct it tonight. Just wait for a night. Thanks $\endgroup$
    – qifeng618
    Commented Apr 22 at 12:42
  • $\begingroup$ @StevenClark Numerical computation shows that the derivative formula $$ f^{(n)}(x)=\frac{f(x)}{2^n \mu^{2n} x^{2n}}\sum_{k=0}^{n}\lambda^k (-2x)^{n-k} \sum_{\ell=0}^{k}\binom{n}{\ell} \binom{n-k-1}{k-\ell-1} \frac{(n-\ell)!}{(k-\ell)!} \mu^{2(n-\ell)} \bigl(\mu^2-x^2\bigr)^\ell $$ for $n\ge0$ is really not correct. But I cannot find any error in the above proof. What's wrong? What stupid mistake did I make in the above proof? $\endgroup$
    – qifeng618
    Commented Apr 23 at 2:56
  • $\begingroup$ In fact, the boxed formula in this answer is correct. $\endgroup$
    – qifeng618
    Commented Apr 24 at 12:20
  • $\begingroup$ No, it is not correct. E.g., $f^{(1)}(x) = (\frac{\lambda}{2x^2}-\frac{\lambda}{2\mu^2})f(x)$, but your formula gives $f^{(1)}(x) = (\frac{\lambda}{x^2}-\frac{\lambda}{2\mu^2})f(x).$ $\endgroup$
    – Fred Hucht
    Commented Apr 24 at 12:53
4
$\begingroup$

We can give a fairly simple generating function for the polynomials $P_{\lambda,\mu}$ from which explicit formulas could be derived.

By Taylor's theorem, $$\sum_{n=0}^\infty f^{(n)}(x) \frac{y^n}{n!} = f(x+y).$$ Writing $P_{\lambda,\mu,n}$ for the proposer's $P_{\lambda,\mu}$, we then have \begin{align*} \sum_{n=0}P_{\lambda,\mu,n}(x) \frac{y^n}{n!} &= f(x+2\mu^2 x^2 y)/f(x)\\ &= \exp\left(\frac{-\lambda y (x^2-\mu^2+2\mu^2x^3y)}{1+2\mu^2xy}\right). \end{align*} So if we define polynomials $Q_n(a,b,c)$ by $$\sum_{n=0}^\infty Q_n(a,b,c)\frac{y^n}{n!} = \exp\left(\frac{ay+by^2}{1+cy}\right)$$ then $$P_{\lambda,\mu,n}(x) = Q_n\bigl(\lambda(\mu^2-x^2), -2\lambda\mu^2 x^3, 2\mu^2 x\bigr).$$

$\endgroup$
2
$\begingroup$

As I mentioned in my Math StackExhange answer Mathematica gives a difference root representation which leads to the formula

$$\frac{\partial^n e^{-\frac{\lambda\, (x-\mu)^2}{2\, \mu^2 x}}}{\partial x^n}=n!\, y(x,n)\tag{1}$$

where $y(x,n)$ is defined recursively as

$$y(x,n)=\left\{\begin{array}{cc} e^{-\frac{\lambda (x-\mu)^2}{2 \mu^2 x}} & n=0 \\ e^{-\frac{\lambda (x-\mu)^2}{2 \mu^2 x}} \frac{\lambda \left( \mu^2-x^2\right)}{2 \mu^2 x^2} & n=1 \\ e^{-\frac{\lambda (x-\mu)^2}{2 \mu^2 x}} \frac{\lambda \left(\lambda \mu^4+\lambda x^4-2 \lambda \mu^2 x^2-4 \mu^4 x\right)}{8 \mu^4 x^4} & n=2 \\ -\frac{\left(-\lambda \mu^2+4 \mu^2 (n-3) x+\lambda x^2+8 \mu^2 x\right)\, y(x,n-1)+2 \left( \mu^2+\mu^2 (n-3)+\lambda x\right)\, y(x,n-2)+\lambda\, y(x,n-3)}{2 \mu^2 n x^2} & n\ge 3 \\ \end{array}\right.\tag{2}$$


WolframAlpha gives the more general result

$$\frac{\partial^n e^{f(x)}}{\partial x^n}=e^{f(x)} \sum\limits_{k=0}^n \frac{1}{k!} \sum\limits_{j=0}^k (-1)^j \binom{k}{j} f(x)^j \frac{\partial ^nf(x)^{k-j}}{\partial x^n}\tag{3}$$

and for

$$f(x)=-\frac{\lambda\, (x-\mu)^2}{2\, \mu^2 x}\tag{4}$$

Mathematica simplifies formula (3) above to

$$\frac{\partial^n e^{-\frac{\lambda\, (x-\mu)^2}{2\, \mu^2 x}}}{\partial x^n}=e^{-\frac{\lambda (x-\mu)^2}{2 \mu^2 x}} n! \sum\limits_{k=0}^n \frac{2^{-k}}{k!} \sum\limits_{j=0}^k (-1)^j \binom{k}{j} \left(-\frac{\lambda (x-\mu)^2}{\mu^2 x}\right)^j y_1(x,n)\tag{5}$$

where $y_1(x,n)$ is defined recursively as

$$y_1(x,n)=\left\{\begin{array}{cc} \left(-\frac{\lambda (x-\mu )^2}{\mu ^2 x}\right)^{k-j} & n=0 \\ \left(-\frac{\lambda (x-\mu )^2}{\mu ^2 x}\right)^{k-j-1} \frac{(\lambda (j-k) (x-\mu ) (\mu +x))}{\mu ^2 x^2} & n=1 \\ -\frac{(\mu (j-k-n+1)+x (j-k+2 (n-1)))\, y_1(x,n-1)+(j-k+n-2)\, y_1(x,n-2)}{n x (x-\mu )} & n\ge 2 \\ \end{array}\right.\tag{6}$$

which is a slightly simpler recursion than formula (2) above.

$\endgroup$
2
$\begingroup$

An alternative answer is \begin{equation*} \frac{\operatorname{d}^n}{\operatorname{d}x^n}\exp\biggl[-\frac{\lambda(x-\mu)^2}{2\mu^2x}\biggr]=\frac{f(x)}{2^n\mu^{2n}x^{2n}}\sum_{k=0}^{n}\binom{n}{k}(-\lambda)^{n-k} 2^{k}\mu^{2k}x^{2(n-k)} \sum_{\ell=0}^k \biggl(-\frac{\lambda}{2}\biggr)^\ell \Biggl[\sum _{j=\ell}^k s(k,j)(-1)^jS(j,\ell)\Biggr] x^{k-\ell} \end{equation*} for $n\ge0$, where $s(n,j)$ and $S(j,k)$ denote the Stirling numbers of the first and second kinds respectively.

References

  1. S. Jin, B.-N. Guo, and F. Qi, Partial Bell polynomials, falling and rising factorials, Stirling numbers, and combinatorial identities, CMES Comput. Model. Eng. Sci. 132 (2022), no. 3, 781--799; available online at https://doi.org/10.32604/cmes.2022.019941.
$\endgroup$
2
  • $\begingroup$ Thank you for this derivation and the link with Stirling numbers! $\endgroup$
    – NancyBoy
    Commented Apr 24 at 7:15
  • $\begingroup$ I have verified this answer. $\endgroup$
    – Fred Hucht
    Commented Apr 24 at 15:46
0
$\begingroup$

In the references below, the derivative formula \begin{equation}\label{exp-frac1x-expans}\tag{QF1} \bigl(\operatorname{e}^{\pm1/t}\bigr)^{(n)} =(-1)^n{\operatorname{e}^{\pm1/t}}\sum_{k=0}^{n}(\pm1)^{k}\binom{n-1}{k-1}\frac{n!}{k!}\frac1{t^{n+k}}, \quad n\ge0 \end{equation} was derived alternatively. From \eqref{exp-frac1x-expans}, it follows that \begin{equation}\label{exp-frac1x-alpha}\tag{QF2} \bigl(\operatorname{e}^{\alpha/t}\bigr)^{(n)} =(-1)^n\operatorname{e}^{\alpha/t}\sum_{k=0}^{n}\alpha^{k}\binom{n-1}{k-1}\frac{n!}{k!}\frac1{t^{n+k}}, \quad n\ge0. \end{equation} Taking $\alpha=-\frac{\lambda}{2}$ in \eqref{exp-frac1x-alpha} gives \begin{equation*} \biggl[\exp\biggl(-\frac{\lambda}{2}\frac{1}{t}\biggr)\biggr]^{(k)} =(-1)^k\exp\biggl(-\frac{\lambda}{2}\frac{1}{t}\biggr) \sum_{\ell=0}^{k}\biggl(-\frac{\lambda}{2}\biggr)^{\ell} \binom{k-1}{\ell-1}\frac{k!}{\ell!}\frac1{t^{k+\ell}}, \quad k\ge0. \end{equation*} Therefore, in light of the Leibnitz rule for differentiation, we obtain \begin{gather*} \begin{aligned} f^{(n)}(x)&=\biggl[\exp\biggl(\frac{\lambda}{\mu}\biggr) \exp\biggl(-\frac{\lambda}{2\mu^2}x\biggr) \exp\biggl(-\frac{\lambda}{2}\frac{1}{x}\biggr)\biggr]^{(n)}\\ &=\exp\biggl(\frac{\lambda}{\mu}\biggr) \sum_{k=0}^{n}\binom{n}{k}\biggl[\exp\biggl(-\frac{\lambda}{2\mu^2}x\biggr)\biggr]^{(n-k)} \biggl[\exp\biggl(-\frac{\lambda}{2}\frac{1}{x}\biggr)\biggr]^{(k)} \end{aligned}\\ \begin{aligned} &=\exp\biggl(\frac{\lambda}{\mu}\biggr) \sum_{k=0}^{n}\binom{n}{k} \exp\biggl(-\frac{\lambda}{2\mu^2}x\biggr) \biggl(-\frac{\lambda}{2\mu^2}\biggr)^{n-k} (-1)^k\exp\biggl(-\frac{\lambda}{2}\frac{1}{x}\biggr) \sum_{\ell=0}^{k}\biggl(-\frac{\lambda}{2}\biggr)^{\ell} \binom{k-1}{\ell-1}\frac{k!}{\ell!}\frac1{x^{k+\ell}}\\ &=\exp\biggl(\frac{\lambda}{\mu}-\frac{\lambda}{2\mu^2}x-\frac{\lambda}{2}\frac{1}{x}\biggr) \sum_{k=0}^{n}\binom{n}{k} \biggl(-\frac{\lambda}{2\mu^2}\biggr)^{n-k} (-1)^k \sum_{\ell=0}^{k}\biggl(-\frac{\lambda}{2}\biggr)^{\ell} \binom{k-1}{\ell-1}\frac{k!}{\ell!}\frac1{x^{k+\ell}}\\ &=f(x)\frac{(-1)^nn!}{2^n\mu^{2n}x^{2n}} \sum_{k=0}^{n}\frac{\lambda^{n-k}}{(n-k)!} 2^{k}\mu^{2k} \sum_{\ell=0}^{k}\frac{(-\lambda)^{\ell}}{(2\ell)!!} \binom{k-1}{\ell-1}x^{2n-k-\ell}, \quad n\ge0. \end{aligned} \end{gather*} In conclusion, we obtain the derivative formula \begin{equation*} \boxed{f^{(n)}(x)=f(x)\frac{(-1)^nn!}{2^n\mu^{2n}x^{2n}} \sum_{k=0}^{n}\frac{\lambda^{n-k}}{(n-k)!} 2^{k}\mu^{2k} \sum_{\ell=0}^{k}\frac{(-\lambda)^{\ell}}{(2\ell)!!} \binom{k-1}{\ell-1}x^{2n-k-\ell}, \quad n\ge0.} \end{equation*}

References

  1. S. Daboul, J. Mangaldan, M. Z. Spivey, and P. J. Taylor, The Lah numbers and the $n$th derivative of $e^{1/x}$, Math. Mag. 86 (2013), no. 1, 39--47; Available online at http://dx.doi.org/10.4169/math.mag.86.1.039.
  2. F. Qi, An explicit formula for the Bell numbers in terms of the Lah and Stirling numbers, Mediterr. J. Math. 13 (2016), no. 5, 2795--2800; available online at https://doi.org/10.1007/s00009-015-0655-7.
  3. X.-J. Zhang, F. Qi, and W.-H. Li, Properties of three functions relating to the exponential function and the existence of partitions of unity, Int. J. Open Probl. Comput. Sci. Math. 5 (2012), no. 3, 122--127; available online at https://doi.org/10.12816/0006128.
  4. https://qifeng618.wordpress.com/2018/05/10/some-papers-related-to-the-function-e1-x-and-the-lah-numbers/
$\endgroup$
4
  • $\begingroup$ Your derivation down to and including your second to last formula $$\frac{\partial^n e^{-\frac{\lambda (x-\mu)^2}{2 \mu^2 x}}}{\partial x^n}=e^{-\frac{\lambda (x-\mu)^2}{2 \mu^2 x}} \sum_{k=0}^{n}\binom{n}{k} \biggl(-\frac{\lambda}{2\mu^2}\biggr)^{n-k} (-1)^k \sum_{\ell=0}^{k}\biggl(-\frac{\lambda}{2}\biggr)^{\ell} \binom{k-1}{\ell-1}\frac{k!}{\ell!}\frac1{x^{k+\ell}}$$ seems to validate, so there must be a mistake in the final step where you moved some of the terms out of the two sums. $\endgroup$ Commented Apr 24 at 0:41
  • $\begingroup$ @StevenClark I don't find any error in my proof. I numerically verified the boxed derivative formula by the software Mathematica 12 and didn't find it wrong again. I don't know what happened before. Consequently, the boxed derivative formula is correct. By the way, I want to know why you voted down this answer. $\endgroup$
    – qifeng618
    Commented Apr 24 at 9:06
  • $\begingroup$ I didn't down-vote your answer, I actually like it because it eliminates the recursion in my answer. I didn't actually try your final formula before since you originally indicated it wasn't correct, but I see now it evaluates correctly. I upvoted to cancel the downvote. $\endgroup$ Commented Apr 24 at 13:26
  • $\begingroup$ @StevenClark I am sorry that I misunderstood your words “ Your derivation down to and including your second to last formula”. Thank you very much for your upvoting this answer. I don’t know why I mistakenly verified the boxed formula several times, perhaps my software Mathematica has bugs or some computer virus intruded into my computer. This time is a lesson for me. Anyway, to the end, we now confirm the boxed formula correct $\endgroup$
    – qifeng618
    Commented Apr 24 at 23:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.