I don't know if this is the proper place for asking this kind of questions and I apologise in advance if it isn't, but anyways: is there a way to propagate linear uncertainties (i.e. through first-degree partial derivatives) for functions in an open form, that is, that cannot be reduced? I am particularly interested in finding the uncertainty of $y$, given the equation:

$y = A\cdot sinh(y)\cdot\sqrt{BC}$

where $A$, $B$, $C$ all have a given uncertainty. Does anybody have any insight? Thank you.

  • $\begingroup$ Could you elaborate on what "open form" and "cannot be reduced" might mean? Would you perhaps mean an implicitly defined function, such as propagating the uncertainty from the vector $(A,B,C)$ to $y$ given that $F(y,A,B,C)=0$ for a known function $F$? $\endgroup$ – whuber May 20 '15 at 19:33
  • $\begingroup$ Did you really intend for $y$ to be on both sides of that equation? Is there any time dependence in there, or are those both the same $y$? $\endgroup$ – Glen_b May 21 '15 at 7:04

Sure. Consider

$f(y) = y/\sinh y = g(A,B,C) = A \sqrt{BC}$


$ \begin{align} \sigma_f = \sigma_y \left| \frac{df}{dy} \right|_{y = \overline{y}} = \sigma_y \mathop{\rm csch} \overline{y} \: (1 - \overline{y} \mathop{\rm \coth \overline{y}}) = \sigma_g \end{align} $

where $\overline{y}$ is the expected value of $y$. You can work out $\sigma_g$ as a function of $\sigma_{A,B,C}$ by the usual error propagation.

Of course, this technique won't work if $\sigma_y$ is so large that $\frac{df}{dy}$ becomes nonlinear within $\overline{y} \pm \sigma_y$.


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.