Hi Sebastien, Nope - Error propagation works according to Gauss like: [sigma(f(a,b,...))]^2= [df/da]^2 * [sigma(a)]^2 + [df/db]^2 * [sigma(b)]^2 + .... In other words the variance (square of the standard deviation) of a function f that depends on variables a,b,... is the sqare of the partial derivative of f with respect to a [1], times the square of the standard deviation of a [2] + the square of the partial derivative of f with respect to b times the square of the standard deviation of b + and so on... For NOE=a/b you get: dNOE/da = 1/b dNOE/db = -a/(b^2) If you insert that into the formula above and take the square root at the end to get sigma(NOE) instead of sigma(NOE)^2 then you end up with Ed's formula. Cheers, Horst Joachim Schirra [1]=(possibly better written in the proper form as [d f(a,b,...)/da]^2 - but I wanted to save space up there) [2]= In other words: the variance of a ---------------------------------------------------------------- / Dr. sc. nat. Horst Joachim Schirra Phone: (+61)7/3346-2021 / / Queensland Smart State Fellow Fax: (+61)7/3346-2101 / / Institute for Molecular Bioscience / / University of Queensland email: h.schirra@xxxxxxxxx / / Brisbane QLD 4072, Australia http://www.uq.edu.au/~uqhschir / ---------------------------------------------------------------- ----- Original Message ----- From: Sebastien Morin <sebastien.morin.1@xxxxxxxxx> Date: Friday, August 17, 2007 3:37 am Subject: NOE errors
Hi everyone, I was looking at the NOE tool in relax and was quite surprised by the way errors are calculated... In the file 'specific_fns/noe.py' of the 1.2 line (r3354), the functionfor the noe error is : _____________________________________________ \/ {sd(sat)*I(unsat)}^2 + {sd(unsat)*I(sat)}^2 sd(NOE) = ---------------------------------------------- - I(unsat)^2 Shouldn't this be more like : _____________________________________________ sd(NOE) = NOE * \/ {sd(sat)*I(sat)}^2 + {sd(unsat)*I(unsat)}^2 In other words, shouldn't the NOE error be the product of the NOE by the root of the squared sum of relative errors for Isat and Iunsat, respectively ? Ain't this the way one should normally propagate errors for value calculated by a division (like NOE=Isat/Iunsat) What are the main advantages of calculating errors with either techniques and why the one is relax should be more accurate ? Thanks ! Sébastien -- ______________________________________ _______________________________________________ | | || Sebastien Morin || ||| Etudiant au PhD en biochimie ||| |||| Laboratoire de resonance magnetique nucleaire |||| ||||| Dr Stephane Gagne ||||| |||| CREFSIP (Universite Laval, Quebec, CANADA) |||| ||| 1-418-656-2131 #4530 ||| || || |_______________________________________________| ______________________________________ _______________________________________________ relax (http://nmr-relax.com) This is the relax-users mailing list relax-users@xxxxxxx To unsubscribe from this list, get a password reminder, or change your subscription options, visit the list information page at https://mail.gna.org/listinfo/relax-users