When considering the uncertainty of a signal which is determined by multiple inputs propagation of uncertainty states that for measurement $$ y = f(x), x=\{x_1, x_2,..., x_N\} $$ uncertainty in $y$ is $$u_y=\sum_i \frac{\partial f}{\partial x_i} u_{x_i} $$ My question is how do we get to this rule. I know that it can be obtained by a taylor expansion of $f_i$ but I don't see why this is a sensible/logical thing to do.

  • 1
    $\begingroup$ Please define uncertainty. What your formula looks like is just a step along the definition of derivative of $y$ with respect to $x$: $$\Delta y = \sum_i \Delta f_i(x) = \sum_i \frac{\Delta f_i}{\Delta x}\Delta x$$ $\endgroup$ Commented May 1, 2014 at 17:24
  • $\begingroup$ @DilipSarwate The uncertainty of a measurement is basically the standard deviation (or a multiple) that you would expect if you repeated the measurement multiple times. See wikipedia for more about propagation $\endgroup$
    – nivag
    Commented May 2, 2014 at 8:25
  • $\begingroup$ Do you actually mean $y(x_1,\ldots,x_n)=\sum_if_i(x_1,\ldots,x_n)$? $\endgroup$
    – Matt L.
    Commented May 2, 2014 at 9:16
  • $\begingroup$ @MattL. Not quite, but I definitely didn't mean what I had written. Edited this now, hopefully it makes more sense $\endgroup$
    – nivag
    Commented May 2, 2014 at 9:25
  • $\begingroup$ If the uncertainty does mean standard deviation as you claim, then your formula is sheer nonsense. It might make sense of uncertainty means variance. $\endgroup$ Commented May 3, 2014 at 1:52


Your Answer

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

Browse other questions tagged or ask your own question.