5

I am trying to find the $n$-th derivative of $\csc(m\pi)$, so I took few cases:

for simplicity let $x=\cot(m\pi)$ and $y=\csc(m\pi)$,

$$\frac{d^0}{dm^0}\csc(m\pi)=\pi^0(\color{red}{1}x^0y^1)$$

$$\frac{d^1}{dm^1}\csc(m\pi)=-\pi^1 (\color{red}{1}x^1y^1)$$

$$\frac{d^2}{dm^2}\csc(m\pi)=\pi^2(\color{red}{1}x^2y^1+\color{red}{1}x^0y^3)$$

$$\frac{d^3}{dm^3}\csc(m\pi)=-\pi^3(\color{red}{1}x^3y^1+\color{red}{5}x^1y^3)$$

$$\frac{d^4}{dm^4}\csc(m\pi)=\pi^4(\color{red}{1}x^4y^1+\color{red}{18}x^2y^3+\color{red}{5}x^0y^5)$$

$$\frac{d^5}{dm^5}\csc(m\pi)=-\pi^5(\color{red}{1}x^5y^1+\color{red}{58}x^3y^3+\color{red}{61}x^1y^5)$$

$$\frac{d^6}{dm^6}\csc(m\pi)=\pi^6(\color{red}{1}x^6y^1+\color{red}{179}x^4y^3+\color{red}{479}x^2y^5+\color{red}{61}x^0y^7)$$

and saw that

\begin{align} \frac{d^n}{dm^n}\csc(m\pi)&=(-\pi)^n\sum_{k=0}^{\lfloor{n/2}\rfloor}\color{red}{a_k} x^{n-2k} y^{2k+1}\\ &=(-\pi)^n\sum_{k=0}^{\lfloor{n/2}\rfloor}\color{red}{a_k} \cot^{n-2k}(m\pi)\csc^{2k+1}(m\pi)\\ &=(-\pi)^n\csc^{n+1}(m\pi)\sum_{k=0}^{\lfloor{n/2}\rfloor}\color{red}{a_k} \cos^{n-2k}(m\pi) \end{align}

If we replace $n$ by $2n$ then separate the last term we have

$$\frac{d^{2n}}{dm^{2n}}\csc(m\pi)=\pi^{2n}\csc^{2n+1}(m\pi)\left[a_n+\sum_{k=0}^{n-1}\color{red}{a_k} \cos^{2n-2k}(m\pi)\right]$$

In the cases mentioned above, we notice that when the order of the derivative is $0, 2, 4, 6$, the coefficients of the last terms are $1, 1, 5, 61$ which match the absolute value of the Euler numbers:

$$E_0=1, E_2=-1, E_4=5, E_6=-61$$

and so

$$\frac{d^{2n}}{dm^{2n}}\csc(m\pi)=\pi^{2n}\csc^{2n+1}(m\pi)\left[|E_{2n}|+\sum_{k=0}^{n-1}\color{red}{a_k} \cos^{2n-2k}(m\pi)\right]$$

By the way, if we take the limit to both sides of the last result letting $m$ approach $1/2$, we have

$$\lim_{m\to \frac12}\frac{d^{2n}}{dm^{2n}}\csc(m\pi)=\pi^{2n}(1)\left[|E_{2n}|+0\right]=|E_{2n}|\pi^{2n}.$$


Question: Is it possible to find $\color{red}{a_k}$?

Ali Shadhar
  • 25,498
  • Perhaps use induction to find a recursive relation for the coefficients – Golden_Ratio Mar 05 '22 at 01:38
  • 1
    My two cents: The coefficients of y^3 are, by induction, equal to (1/4)(3^n-(2n+1)) while the coefficients of y^5 are, again by induction, equal to (1/16)(5^n-3^n(2n-1)+2(n^2-n-1)). Found on the OEIS entries A000340 and A000363 respectively. Hope this helps. – Michael Smith Mar 06 '22 at 06:36
  • @Michael Smith thank you Michael, can you show some details in solution? – Ali Shadhar Mar 06 '22 at 06:39
  • Sure i do not mind, though I don't think I have much else to say on this without spending time searching through the coefficients of higher order terms. What details would you want me to provide? i'd be glad to – Michael Smith Mar 06 '22 at 06:43
  • I just came across something that might be even more helpful. In the example section of this OEIS entry https://oeis.org/A008971 there is a triangle with the same coefficients, that seems to be related to some combinatorial results apparently. – Michael Smith Mar 06 '22 at 06:50
  • @Michael Smith I would be glad to see you succeed in solving the problem using the reference you provided. – Ali Shadhar Mar 06 '22 at 06:52
  • Actually, I'd be glad as well, though i do not see how to get an explicit formula using the generating functions provided at A008971. I'm optimistic that someone will pick this up where I left off. I will try once more to crack this later if noone has beaten me to it and see if I have any better luck. – Michael Smith Mar 06 '22 at 06:56
  • 1
    @@Michael Smith that's fine and thank you for participating I appreciate your trying it – Ali Shadhar Mar 06 '22 at 06:58

1 Answers1

3

Thanks to @Domen for his solution, the answer is

$$\frac{d^n}{dm^n}\csc(m\pi)=(-\pi)^n\csc^{n+1}(m\pi)\sum_{k=0}^{\lfloor{n/2}\rfloor}t(n,k) \cos^{n-2k}(m\pi)$$

where

$$t(n,k)=(2k+1)t(n-1,k)+(n-2k+1)t(n-1,k-1)$$

and

$$t(n,0)=1$$


Different form:

The $n$-th derivative of $\sec(x)$ is given by Wolfram:

$$\frac{d^{ n}}{d{x}^{n}}\sec(x)= \sum_{k=0}^{\infty} \frac{|E_{2 k}|}{(2 k-n) !} x^{2 k-n}$$

Proof:

By Taylor series we have

$$\sec(x)=1+\frac1{2!}x^2+\frac{5}{4!}x^4+\frac{61}{6!}x^6+....=\sum_{k=0}^\infty\frac{|E_{2k}|}{(2k)!}x^{2k}$$

Take the $n$-th derivative to both sides

$$ \frac {d^{ n}}{d{x}^{n}}\sec(x)=\sum_{k=0}^\infty\frac{|E_{2k}|}{(2k)!} \frac {d^{ n}}{d{x}^{n}}x^{2k}$$

We have

$$\frac {d}{dx}x^{2k}=2k x^{2k-1}$$

$$\frac {d^{ 2}}{d{x}^{2}}x^{2k}=2k(2k-1) x^{2k-2}$$

$$\frac {d^{ 3}}{d{x}^{3}}x^{2k}=2k(2k-1)(2k-2) x^{2k-3}$$

note that

$$2k(2k-1)(2k-2)*\color{red}{\frac{(2k-3)(2k-4)...}{(2k-3)(2k-4)...}}=\frac{(2k)!}{(2k-3)!}$$

so in general we have

$$\frac {d^{ n}}{d{x}^{n}}x^{2k}=\frac{(2k)!}{(2k-n)!}x^{2k-n}$$

Thus,

$$ \frac {d^{ n}}{d{x}^{n}}\sec(x)= \sum_{k=0}^{\infty} \frac{|E_{2 k}|}{(2 k-n) !} x^{2 k-n}$$

Finally, by using $\sec(x-\pi/2)=\csc(x)$ we have

$$ \boxed{\frac {d^{ n}}{d{x}^{n}}\csc(x)= \sum_{k=0}^{\infty} \frac{|E_{2 k}|}{(2 k-n) !} \left(x-\frac{\pi}{2}\right)^{2 k-n}}$$


Edit. A more rigorous proof: Let $z=x+1/2$,

\begin{gather*} \lim_{z\to \frac12}\frac{d^{2a}}{d z^{2a}}\csc(z\pi)=\lim_{x\to 0}\frac{d^{2a}}{d x^{2a}}\csc\left(x\pi+\frac{\pi}{2}\right)\\ =\lim_{x\to 0}\frac{d^{2a}}{d x^{2a}}\sec(x\pi)\\ \left\{\text{expand $\sec(x\pi)$ in series}\right\}\\ =\lim_{x\to 0}\frac{d^{2a}}{d x^{2a}}\sum_{k=0}^\infty \underbrace{\frac{|E_{2k}|\pi^{2k}}{(2k)!}}_{f_{2k}}x^{2k}\\ =\lim_{x\to 0}\frac{d^{2a}}{dx^{2a}}\sum_{k=0}^\infty f_{2k}\,x^{2k}\\ =\lim_{x\to 0}\frac{d^{2a}}{dx^{2a}}\left(f_0x^0+f_2x^2+f_4x^4+...\right)\\ =(2a)! f_{2a}\\ =(2a)!\frac{|E_{2a}|\pi^{2a}}{(2a)!}\\ =|E_{2a}|\pi^{2a}. \end{gather*}

Ali Shadhar
  • 25,498
  • 1
    Awesome solution, didn't know mathematica could do something like this, though want to mention that this was also directly below the egf in the formula section of this oeis entry https://oeis.org/A008971 directly below the egf – Michael Smith Mar 09 '22 at 01:18
  • 1
    See also https://functions.wolfram.com/ElementaryFunctions/Csc/20/02/0004/ – Gary Mar 09 '22 at 02:15
  • Yes @Michael Smith i see it in the link. It would be much nicer if the function of $a_n$ is given in explicit form not recursive one. – Ali Shadhar Mar 09 '22 at 02:41
  • Thank you @Gary thats an amazing reference. Are you familiar with the proof? – Ali Shadhar Mar 09 '22 at 03:38
  • 1
    @MichaelSmith I discovered recently that Maple can do that too, i.e. it can simplify an expression like diff(sec(x),x$n), in this case using a sum involving Stirling numbers of the second kind. – Jean-Claude Arbaut Mar 13 '22 at 03:00