NavList:
A Community Devoted to the Preservation and Practice of Celestial Navigation and Other Methods of Traditional Wayfinding
Re: Averaging lunars: was Lunars with SNO-T
From: Herbert Prinz
Date: 2004 Nov 1, 05:52 -0500
From: Herbert Prinz
Date: 2004 Nov 1, 05:52 -0500
George Huxtable wrote: > It may be interesting to read what Tobias Mayer had to say on this topic, > [...] Thanks to George Huxtable for posting and Steven Wepster for translating. While we are at it, Note I, immediately preceding the quoted passage might also be of interest. In an earlier message I had casually compared the repeating circle with our averaging technique. This note reminds me that the two methods are different in one respect. In Note I, Mayer says that the final error of an observation is inverse proportional to the number of individual distances that have been accumulated into this observation. Therefore, if one assumes an instrument error of 2' to 3', six observations are plenty sufficient to get the error below 30". In passing I remark that the goal for accuracy that Mayer sets himself is truly down to earth, much less ambitious than that of some of our list members! But my question is: Is his arithmetic right? It is right, if we assume that the observer is perfect and the only purpose of the averaging process is the elimination of instrument error and reading error (due to limited resolution of the division of the limb). This is what Mayer had in mind. When we average with a sextant, we are in a different situation. We can never get around the instrument error or the reading error with repeated readings. Our assumption is that the observer is not perfect and each observation is fraught with random error. This error is inverse proportional to the square root of the number of observations. Given the above parameters, six observations would not be sufficient. If an imperfect observer producing random errors had a perfect sextant and a perfect repeating circle, the result of a repeated observation would be the same on both instruments (statistically). The error would obey the inverse square root law on both instruments. Mayer did not address this kind of error. He did not have the appropriate error theory. Might he have thought that he had divided random error by the number of observations, too? I don't know the answer. Herbert Prinz