2
$\begingroup$

I have the below function:

$$\pi(x) = \frac{s_0\cdot \left(1-\left(\frac{s_1}{s_1+x \cdot \lambda}\right)^{k}\right) \cdot r_1}{s_0\cdot \left(1-\left(\frac{s_1}{s_1+x \cdot \lambda}\right)^{k}\right) \cdot \gamma+ r_0}- x$$ with:

  • $x, s_0, s_1, r_0, r_1 \in R^+$
  • $\lambda, \gamma \in [0,1]$
  • $k \in \{0.25, 4\}$

I am trying to find the $x$ that maximises $\pi(x)$ for both k values (by graphing the function on Desmos, one can see there exists a maximum).

We solve $\frac{d \pi(x)}{dx} = 0$ i.e. $$ \frac{\left(s_{0}\cdot r_{1}\cdot r_{0}\cdot\lambda\cdot k\cdot\left(\frac{s_{1}}{s_{1}+\lambda\cdot x}\right)^{1+k}\right)}{s_{1}\cdot\left(s_{0}\cdot\gamma\cdot\left(\left(\frac{s_{1}}{s_{1}+\lambda\cdot x}\right)^{k}-1\right)-r_{0}\right)^{2}}-1=0$$

$$\left(s_{0}\cdot r_{1}\cdot r_{0}\cdot\lambda\cdot k\cdot\left(\frac{s_{1}}{s_{1}+\lambda\cdot x}\right)^{1+k}\right) = s_{1}\cdot\left(s_{0}\cdot\gamma\cdot\left(\left(\frac{s_{1}}{s_{1}+\lambda\cdot x}\right)^{k}-1\right)-r_{0}\right)^{2}$$

We introduce:

  • $a=s_{0}\cdot r_{1}\cdot r_{0}\cdot\lambda \cdot k$
  • $y=s_1+\lambda \cdot x$

and rewrite $$ \pm\sqrt{a} \cdot s_1^{\frac{k}{2}} \cdot y^{-\frac{k+1}{2}} = s_0 \cdot \gamma \cdot \left(\left(\frac{s_{1}}{y}\right)^{k}-1\right)-r_{0} $$ $$ \pm \sqrt{a} \cdot s_1^{\frac{k}{2}} \cdot y^{-\frac{k+1}{2}} - s_0 \cdot \gamma \cdot s_{1}^k \cdot y^{-k} + s_0 \cdot \gamma +r_{0} =0 $$

and going one step further:

$$A \cdot y^{-\frac{k+1}{2}} + B \cdot y^{-k} + C= 0$$ with:

  • $A=\pm \sqrt{s_{0} \cdot r_{1} \cdot r_{0} \cdot \lambda \cdot k \cdot s_1^k}$
  • $B = - s_0 \cdot \gamma \cdot s_{1}^k$
  • $C=s_0 \cdot \gamma +r_{0}$

How can we go from here? One solution to make things easier could be to rewrite the equation with $t=\frac{1}{\sqrt{y}}$ but not sure that leads anywhere.

NB: this is a repost from maths.stackexchange as it seems to be more appropriate for mathoverflow as more research oriented

$\endgroup$
3
  • $\begingroup$ mathoverflow.net/questions/418392/… $\endgroup$ Commented Mar 25, 2024 at 9:38
  • $\begingroup$ @CaveJohnson Does that work for a pseudo trinomial in the sense that here the exponents being negative, this is not a real trinomial? $\endgroup$ Commented Mar 25, 2024 at 9:43
  • $\begingroup$ Hi Arthur! I think you wrote to me on April 15 asking for this problem, and I sent you a more or less complete solution in terms of hypergeometric series, and by Newton iteration for comparison. Was it ok? $\endgroup$ Commented Aug 23, 2024 at 13:30

1 Answer 1

0
$\begingroup$

After spending some time on it, below is my proposed solution.

We can undergo another change of variables to make it a trinomial: $t=\frac{1}{\sqrt{y}}$ and we obtain: \begin{align} A \cdot t^{k+1} + B \cdot t^{2k} +C =0 \ \ (1) \end{align}

Case of k=4

For $k=4$, the exponents $k+1=5$ and $2k=8$ are integers, making (1) a proper trinomial: \begin{align} A \cdot t^{5} + B \cdot t^{8} +C =0 \ \ (2) \end{align}

There is probably a Lagrange inversion theorem series solution to it, we try to formalize it: The problem has the form $z=f(w)$ with:

  • $f(w)=A \cdot w^5+B \cdot w^8$
  • $f'(w) = 5A\cdot w^4+8B\cdot w^7$
  • $z=-C$

$f$ is analytical in $w_0=1$ and $f'(w_0)=5A+8B \neq 0$.

We can now express the form $w=h(z)$ with: $$h(z) = w_0 + \sum_{n=1}^\infty h_n \cdot \frac{\left(z-f(w_0)\right)^n}{n!}$$

where:

$$h_n=\lim_{w \rightarrow w_0} \frac{d^{n-1}}{dw^{n-1}}\left[\left(\frac{w-w_0}{f(w) - f(w_0)}\right)^n\right]$$

Case of k=1/4

For $k=0.25$, the exponents are $k+1=1.25$ and $2k=0.5$ which are not integers, so we could make another change of variable so that solving (1) is equivalent to solving: \begin{align} A \cdot t^{5} + B \cdot t^{2} +C =0 \ \ (3) \end{align}

And we can follow a similar reasoning as in the case of $k=4$ to find a Lagrange inversion theorem series solution

$\endgroup$

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.