0
$\begingroup$

I have a question, here it is -

Given $Y = \sin(\theta)$, find percentage error in $y$ if percentage error in $\theta$ is $2\%$ at $\theta = \dfrac{\pi}{6} \ \text{rad}$.

Our teacher did this solution -
$$y = \sin \theta.$$
Differentiate $y$ wrt $\theta$
$$\frac{\mathrm{d}y}{\mathrm{d}\theta} = \cos \theta \iff \mathrm{d}y = \cos \theta \, \mathrm{d}\theta.$$
Divide both sides by $y$:
$$\frac{\mathrm{d}y}{y} = \frac{\cos \theta \, \mathrm{d}\theta}{\sin \theta}.$$
Multiply by $100$ on both sides
$$\left(\frac{\mathrm{d}y}{y}\right) \times 100 = \frac{\cos \theta \, \mathrm{d}\theta}{\sin \theta} \times 100.$$
So: $$ \% \ \text{error in} \, y = \cot \theta \ \mathrm{d}\theta \times 100.$$
Multiply and divide by $\theta$ on RHS:
$$ \% \ \text{error in} \, y = \frac{\cot \theta \ \mathrm{d}\theta \cdot \theta}{\theta} \times 100.$$
Putting the values of $$ \frac{\mathrm{d}\theta}{\theta} \times 100 = 2 \% \quad \text{and} \quad \theta = \frac{\pi}{6} \ \text{rad},$$
we have $$\begin{align} \% \ \text{error in} \, y &= \cot \frac{\pi}{6} \times 2 \times \frac{\pi}{6} \\ &= \sqrt 3 \times \frac{\pi}{3} \\ &= \frac{\pi}{\sqrt 3} \ \% \\ &\approx 1.81 \%. \end{align}$$

But I did this:
$$y = \sin \theta$$
$$y_t = \sin \frac{\pi}{6} = \frac{1}{2} = 0.5,$$ where $y_t$ means true value of $y$.
Assuming the case of positive error:

$$\begin{align} y_m &= \sin \frac{\pi}{6} + \frac{2}{100} \times \frac{\pi}{6} \\ &= \sin \frac{\pi}{6} \cos \frac{\pi}{300} + \cos \frac{\pi}{6} \sin \frac{\pi}{300} \\ &\approx 1. \end{align}\\ $$ \begin{align} \therefore \, \% \ \text{error in} \, y &= \frac{|y_t - y_m|}{y_t} \times 100 \\ &= \frac{|0.5 - 1|}{0.5} \times 100 \\ &= 100\%, \end{align}

where $y_m$ means measured value of $y$.

I want to know what mistake I made or why my approach is wrong. Thank you.

$\endgroup$

1 Answer 1

3
$\begingroup$

The error seems to mostly be in your evaluation of $\sin\left(\frac{\pi}{6}\right)\cos\left(\frac{\pi}{300}\right)+\cos\left(\frac{\pi}{6}\right)\sin\left(\frac{\pi}{300}\right)$, which I find equals approximately $0.509$. You can tell the result should be close to $0.5$ and not close to $1$ like your result because $\cos(\pi/300)\approx 1$ and $\sin(\pi/300)\approx 0$ so the result should be $\approx \sin(\pi/6)=0.5$. The percent error $|0.5-0.509|/0.5\approx 1.81\%$ agrees.

Then the question of whether your method should or should not give the same result requires a slightly more sophisticated discussion of the origin of the formula for error propagation. You assumed a positive error, but could have found a different result for a negative error, or the average of the positive and negative error, etc, depending on the functions at hand. There is also a question of the distribution of the error (is there an equal chance of every value from 0 to 2%, or is it more likely to be 0% and there's actually some small chance of errors higher than 2%?). Everything should agree when the errors are small, fortunately.

$\endgroup$
3
  • $\begingroup$ Thanks for answering! A quick clarification: since the conventional method of error propagation fails to solve for bigger errors like 50%, we were taught to use different methods for errors less than or equal to 10% and greater than that. So, will both ways work out for minute errors? $\endgroup$ Commented Oct 8 at 19:29
  • $\begingroup$ @ApogeePoint yes - see en.wikipedia.org/wiki/Propagation_of_uncertainty#Example for example. Linearizations of functions tend to be pretty accurate close to some value but, depending on the function, they deviate from the true value when you go too far (if you draw a tangent to a sinusoidal curve, it follows the curve for a bit but then deviates away) $\endgroup$ Commented Oct 8 at 21:13
  • 1
    $\begingroup$ The mention of linearisation of functions cleared it all, kudos to you! $\endgroup$ Commented Oct 8 at 23:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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