4

$$\frac{d^n}{dx^n}(f(x))^n \text{ and similar expressions}$$


I was trying to work out the Taylor series for the solution to the general cubic using Lagrange Inversion (out of curiosity: I was wondering which of the roots the series would give for different cubics) and got stuck on finding the $n^{th}$ derivative of $ (\frac{1}{ax^2+bx+c})^{n+1} $.

First I tried working it out by hand but I couldn't see any obvious pattern that would allow me to compute the general case. Wolfram Alpha was able to calculate the next few values of n and I noticed some things but there was no obvious pattern for the general term.

I realized I could do it by employing the Laplace Transform (or the Fourier Transform or maybe even directly from the Taylor Series) where taking n+1 derivatives is trivial but found it rather difficult to compute any of these in general since the powers of $ \frac{1}{ax^2+bx+c}$ are rather chaotic. I am a bit stuck on this.

Also, if someone knows how to do this, I'm now curious if there is a way to get the general ${n-1}^{th}$, $n^{th}$, or ${n+1}^{th}$ derivatives of the $n^{th}$ or ${g(n)}^{th}$ powers of a function f(x) for some function g(n), and as to where the coefficients of these terms come from (some sort of analogue of Pascal's or one of Stirling's triangles maybe?).

For example, the coefficients of the $n^{th}$ derivatives of $(f(x))^n$ can be arranged in a triangle like this when you arrange them in increasing order in each row:

$$\text{1}$$

$$\text{2 2}$$

$$\text{3 6 18}$$

$$\text{4 24 36 48 144}$$

But I don't remember seeing these numbers anywhere and couldn't even find this sequence (read out by rows) in the OEIS.

TheJack
  • 492
  • 1
    Instead of your derivative setup, try Lagrage reversion to get $ax^2+bx+c=0\iff x=-\frac cb -\frac a b x^2 \implies x=-\frac c b+\sum\limits_{n=1}^\infty \left(-\frac ab \right)^n\frac1{n!}\left.\frac{d^{n-1}}{dx^{n-1}}(x^2)^n\right|_{x=-\frac cb}$. Also, you could maybe use infinite binomial series twice for $\left(\frac1{ax^2+bx+c}\right)^{n+1}$ instead of the generalized product rule. – Тyma Gaidash Aug 15 '23 at 00:09
  • Never heard of Lagrange Reversion. Thanks for the suggestion – TheJack Aug 15 '23 at 00:14
  • Will this give the same series and converge on the same root? – TheJack Aug 15 '23 at 00:15
  • 1
    It would give a simpler series. The comment’s series would give a real root as in your question, but it may have a different radius of convergence. – Тyma Gaidash Aug 15 '23 at 00:18
  • I see. I will try it out possibly with complex coefficients too....mainly curious because I was wondering if there was some Newtons fractal for Langrange Inversion – TheJack Aug 15 '23 at 00:20
  • 1
    It should work with complex coefficients. If other complex roots are wanted, the cubic equation can be transformed and Lagrange reversion/inversion would give some – Тyma Gaidash Aug 15 '23 at 00:21

1 Answers1

2

There is the broader generalized product rule which states $$ (f_1 \cdot \ldots\cdot f_m)^{(n)} = \sum_{k_1+k_2+\cdots+k_m=n} {n \choose k_1, k_2, \ldots, k_m} \prod_{1\le t\le m}f_{t}^{(k_{t})} $$ Take $m=n$ and $f_i = f$ for all $i$ to get $$ \frac{d^n f^n}{dx^n} = \sum_{k_1 + \cdots + k_n = n} {n \choose k_1, k_2, \ldots, k_n} \prod_{j=1}^n \frac{d^{k_j} f}{dx} $$ These coefficients form the multinomial coefficients, a generalization of the binomial coefficients, which arise from the expansion of $(x+y)^n$, whereas $$ \binom{n}{k_1,\cdots,k_m} := \frac{n!}{k_1! \cdots k_m!} $$ arises from the expansion of $(x_1+\cdots+x_m)^n$.

One can tackle the order $n \pm 1$ cases in the obvious way from here. And I guess order $g(n)$ but I imagine the expression there would range on the nigh useless considering this barely feels adequate. But maybe someone knows something more clever.

PrincessEev
  • 50,606
  • Thanks for the answer! Now I just need to compute the general derivative of the reciprocal of a quadratic probably with PFD. Also I am wondering is there an analogous formula like this for the finite difference operator (these could then be used for FDM)? – TheJack Aug 14 '23 at 22:44