5
$\begingroup$

I am working on an Optimization problem and I would like to suggest finding the solution (i.e. the extrema) of the function below using two methods: the first one being calculating the derivative and making it equal to zero and the second one using a numerical method.

The equation of the model is shown below:

\begin{align*} y = (x^3 - 4x) \cdot \frac{\sqrt{13+4x^2}}{2} \end{align*}

(the derivative being the following)

\begin{align*} y' = \frac{16x^4 +7x^2 - 52}{2 \sqrt{13+4x^2}} \end{align*}

I have two questions about using it:

  • Based on some research, I found it is possible to use Newton's Method on functions that are not polynomials. Is that correct?
  • Is the Newton's Method the most adequate in this case?
$\endgroup$
  • $\begingroup$ Have you got the first derivative? $\endgroup$ – Dr. Sonnhard Graubner Oct 31 '18 at 10:28
  • $\begingroup$ @Dr.SonnhardGraubner yes, that was calculated algebraically (regular product + chain rules) $\endgroup$ – bru1987 Oct 31 '18 at 10:29
  • $\begingroup$ Then Show us your derivative, i'm not sure that you will Need Newton $\endgroup$ – Dr. Sonnhard Graubner Oct 31 '18 at 10:30
  • $\begingroup$ @Dr.SonnhardGraubner there you go. One thing: I deliberately want to use a numerical method for this (even if making the derivative equal to zero was easy) $\endgroup$ – bru1987 Oct 31 '18 at 10:34
  • $\begingroup$ Check your result $\endgroup$ – Dr. Sonnhard Graubner Oct 31 '18 at 10:35
4
$\begingroup$

We want to find the extrema of the function

$$\begin{align*} y(x) = (x^3 - 4x) \cdot \frac{\sqrt{13+4x^2}}{2} \end{align*}$$

One approach is to use Newton's Method on $y'(x)$, where

$$y'(x) = \frac{16 x^4+7 x^2-52}{2 \sqrt{4 x^2+13}}$$

We start by plotting $y(x)$ and see two extrema to try and find using Newton's enter image description here

The iteration formula for Newton's Method is given by

$$x_{n+1} = x_n-\dfrac {f(x_n)}{f'(x_n)} = x_n - \dfrac{16x_n^4 + 7x_n^2 - 52}{2 \sqrt{4 x_n^2 + 13}\left( \dfrac{64 x_n^3 + 14 x_n}{2 \sqrt{4 x_n^2 +13}} - \dfrac{2 x_n (16 x_n^4 + 7 x_n^2 - 52)}{(4 x_n^2 + 13)^{3/2}}\right) }$$

This can be simplified to remove the square roots and nasty divisions as

$$x_{n+1} = x_n - \dfrac{\left(4 x^2+13\right) \left(16 x^4+7 x^2-52\right)}{2 x \left(96 x^4+430 x^2+195\right)}$$

If we initialize Newton's method using $x_0 = -1.4$, we arrive at $x_4 = -1.26382$ in $4$ steps.

If we initialize Newton's method using $x_0 = 1.4$, we arrive at $x_4 = 1.26382$ in $4$ steps.

Update

There is an easier approach to find where $y'(x) = 0$, we need only find the roots of the numerator of $y'(x)$, so the Newton iteration can be simplified to

$$x_{n + 1} = x_n - \dfrac{16 x_n^4 + 7 x_n^2 - 52}{64 x_n^3 + 14 x_n}$$

This gives the same results as before. Also, comparing this with the above iteration, the quadratic term $(4 x^2 + 13)$, leads to imaginary roots that we don't care about.

Lastly, it is worth noting that you can use a simple transform, $t = x^2$, on the iterations' numerator to get $16 t^2 + 7 t - 52 = 0$, and then solve a quadratic formula and eliminate Newton's Method altogether!

$\endgroup$
  • $\begingroup$ NB: you can simplify the denominator of the equation so that you don't need to compute square roots. $\endgroup$ – Jam Oct 31 '18 at 13:20
  • $\begingroup$ @Jam: Good point. I added your comment to the answer, but also added an easier approach using Newton's and a way to eliminate Newton's altogether. $\endgroup$ – Moo Oct 31 '18 at 17:36
-1
$\begingroup$

Hint: I have got this here $$f'(x)={\frac {6\,{x}^{3}-16\,{x}^{2}+13\,x-26}{\sqrt {4\,{x}^{2}+13}}}$$

$\endgroup$
  • $\begingroup$ I just fixed my derivative (double checked on Wolfram). Could you please help me with the questions posted? $\endgroup$ – bru1987 Oct 31 '18 at 11:02
  • $\begingroup$ What want you to solve now? $\endgroup$ – Dr. Sonnhard Graubner Oct 31 '18 at 11:05
  • $\begingroup$ $f' ( x ) = 0 $ with newton method $\endgroup$ – Surb Oct 31 '18 at 11:06
  • $\begingroup$ see here keisan.casio.com/exec/system/1244946907 $\endgroup$ – Dr. Sonnhard Graubner Oct 31 '18 at 11:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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