4
$\begingroup$

I would like to approximate the positive root of the following equation $$ 2(p+1)x^p - px - 2 = 0 $$ where $p$ is an integer. We could use the formula $(1 - y)^p \approx 1 - py$ for $y$ small to get an approximation of root $x_0 \approx \frac{1}{2p+1}$. However, I believe that we can make a stronger approximation. Could you please suggest some ideas?

$\endgroup$
  • 1
    $\begingroup$ Have you tried Newton's method ? (en.wikipedia.org/wiki/Newton%27s_method) $\endgroup$ – user10676 Jun 16 '11 at 8:16
  • $\begingroup$ That approximation doesn't seem to be accurate for large $p$. Did you mean $x_0 \approx 1 - \frac{1}{2p+1}$? $\endgroup$ – Qiaochu Yuan Jun 16 '11 at 10:42
  • 1
    $\begingroup$ If you are going to use this as part of a program, it could be a very good idea to use different "formulas" for the initial approximation for various ranges of values of $p$, say small, medium, and large. $\endgroup$ – André Nicolas Jun 16 '11 at 11:38
13
$\begingroup$

Write $x = 1 - \frac{r}{p}$. Then $x^p \approx e^{-r}$ and the equation becomes (approximately)

$$2(p+1) e^{-r} + r = p+2.$$

For large $p$ the rough approximation $r \approx 1$, substituted into the above, gives $r \approx \ln 2$. Letting $r = \ln 2 + s$ we get

$$(p+1) e^{-s} + \ln 2 + s = (p+2).$$

Since we know $s$ will be small for large $p$ we can further approximate this by

$$(p+1)(1 - s) + \ln 2 + s = (p+2)$$

giving $s = \frac{\ln 2 - 1}{p}$, hence

$$x \approx 1 - \frac{\ln 2}{p} + \frac{1 - \ln 2}{p^2}.$$

I am pretty sure this is accurate to slightly better than second-order:

  • For $p = 10$ the positive root is about $0.936$ and the above gives about $0.934$.
  • For $p = 100$ the positive root is about $0.993123$ and the above gives about $0.993099$.
$\endgroup$

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.