8
$\begingroup$

I have come across a few papers using the notation $A^{\times{B}}_{\div{C}}$; where A,B,C are real number. For example, $3000^{\times{3}}_{\div{4}}$.

An example can be Eqn (4) in Stern & Laor (2012) "Type 1 AGN at low $z$ – II. The relative strength of narrow lines and the nature of intermediate type AGN", which reads as follows:

$$\frac{L_{\rm bol}}{10^{43}\ \rm erg\ s^{-1}} = 4000^{\times 4}_{\div 4} \left( \frac{L_{\rm [O\ III]}}{10^{43}\ \rm erg\ s^{-1}}\right)^{1.39}$$

What is the generic meaning of this notation?

$\endgroup$
1
5
$\begingroup$

Small quibble to the (rightfully) accepted answer by James K that was too long for a comment:


To be fair, $x=24^{+1}_{-3}$ doesn't mean that $21 \le x \le 25$, but that with a particular amount of certainty (usually 68%), $21 \le x \le 25$.

Correspondingly, $x=24^{\times 2}_{\div3}$ would mean that, with some certainty, $8 \le x \le 48$.

Symmetric vs. asymmetric uncertainties

With non-Gaussian, asymmetric errors, given only the two values for the lower and upper error there's no way to know the corresponding 95% interval, 99% interval, and so on. You would have to know the full PDF for that. But if the errors are Gaussian, the $n$'th sigma is equal to $n\sigma$. That is, if the quoted error represents one standard deviation, then for $x=24\pm2$ you know that with 99% certainty the result is $20 \le x \le 28$.

By analogy, if the errors of this notation are normally distributed in log space, as I would think is the case, the $n$'th sigma would be equal to $\sigma^n$. That is, if $x=4000^{\times}_{\div}4$, then $$ \begin{array}{rcl} 1000 \le x\le \phantom{1}16\,000 & (\mathrm{68\% \,\,confidence})\\ \phantom{1}250 \le x\le \phantom{1}64\,000 & (\mathrm{95\% \,\,confidence})\\ \phantom{10}60 \lesssim x\le 256\,000 & (\mathrm{99\% \,\,confidence})\\ (\mathrm{etc.}) & \end{array} $$

Please use logarithms

Personally, I think this notation is horrible. To avoid confusion, instead of $x=4000^{\times}_{\div}4$ I'd much rather write $\log x = 3.6\pm0.6$. Then $$ \begin{array}{rcl} 3.0 \le \log x \le 4.2 & (\mathrm{68\% \,\,confidence})\\ 2.4 \le \log x \le 4.8 & (\mathrm{95\% \,\,confidence})\\ 1.8 \le \log x \le 5.4 & (\mathrm{99\% \,\,confidence})\\ (\mathrm{etc.}) & \end{array} $$ which is (roughly) the same as above.

$\endgroup$
2
  • 1
    $\begingroup$ Thanks, that is wonderfully explained. I am still quite new to this, and if I ever have to write such an equation, I will definitely follow the logarithmic procedure. $\endgroup$ – lordparthurnaax Sep 10 '20 at 9:05
  • 3
    $\begingroup$ This is neither small nor a quibble. $\endgroup$ – uhoh Sep 10 '20 at 10:49
6
$\begingroup$

A confirmation of the explanation is given slightly further on where it talks about how the uncertainty is derived:

In equation (4) we use the 0.6 dex scatter of $\log L_{\rm UV}$ around the $L_{\rm UV}$ versus $L_{\rm [O\ III]}$ relation, as an estimate for the uncertainty in deriving $L_{\rm bol}$ from $L_{\rm [O\ III]}$.

The term "dex" indicates a base-10 logarithm of a unit, which translates into a factor of $10^{0.6} \approx 4$. So translating the error expressed in the logarithm of the quantity into an error expressed on the quantity itself, this becomes "multiply-or-divide by 4", which here is rendered $4000^{\times 4}_{\div 4}$.

Pela's excellent answer explicitly goes into detail about what this implies for the probability distribution. I usually prefer notation like $10^{x \pm \sigma}$ which sticks to more familiar notation and explicitly implies something vaguely log-normal rather than normal, although perhaps it gets a bit cramped when asymmetric error bars come into play.

$\endgroup$
4
  • 3
    $\begingroup$ +1, but I've never seen this notation and I think it's horrible! $\endgroup$ – pela Sep 9 '20 at 19:11
  • 3
    $\begingroup$ @pela - I'm not entirely sold on it myself... I think I prefer notation like $10^{x \pm \sigma}$ instead! $\endgroup$ – user24157 Sep 9 '20 at 19:27
  • $\begingroup$ @pela I have never encountered these notations in any books or classes, but these papers treat them as trivial. I wish they would add a concise note making it clear, or use a more intuitive one, like the one antispinwards suggested. $\endgroup$ – lordparthurnaax Sep 10 '20 at 4:40
  • $\begingroup$ A notation like shown with + and - like that giving errors is common when it is not a symmetric error expressed as +- $\endgroup$ – planetmaker Sep 10 '20 at 5:18
3
$\begingroup$

A notation like $x=24^{+1}_{-3}$ is quite common, it means $24-3<x<24+1$ with a best estimate of 24, and is a way of indicating uncertainty.

The example you give is less common, by analogy $x=24^{\times 2}_{\div3}$ means $ 24\div3 < x < 24\times 2$ ie $x$ is between 8 and 48, with a best estimate of 24

$\endgroup$
0

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.