III .
" On the Proof of the Law of Errors of Observations .
" By M. W. CROFTON , F.R.S. Received March 24 , 1869 .
( Abstract .
) The object of this Paper is to give the mathematical proof , in its most general form , of the law of single errors of observations , on the hypothesis that each error in practice arises from the joint operation of a large number of independent sources of error , each of which , did it exist alone , would occasion errors of extremely small amount as compared generally with those actually produced by all the sources combined .
This proof is contained in a process given for a different object , namely , Poisson 's generalization of Laplace 's investigation of the law of the mean results of a large number of observations , to be found in the 'Connaissance des Temps ' for 1827 , and also in his ' Recherches sir la Probabilite des Jugements ; ' it is also reproduced in Mr. Todhunter 's able ' History of the Theory of Probability .
' It is not therefore pretended that any new results are arrived at in the present Paper .
Considering , however , the importance and celebrity of the question , and the refined and difficult character of Poisson 's analysis , it will not probably be deemed superfluous to show how the same law may be demonstrated with equal generality , in a much more simple and elementary manner .
The difficulty of the general proof seems indeed to have been so extensively felt , that several attempts have been made to simplify it .
However , so far as the present writer is aware , no proof has been given , except Poisson 's , which is not open to grave objection , as based upon unjustifiable assumptions , or as unduly limiting the generality of the investigation .
The mathematical reasoning in this Paper is based entirely on the abovementioned hypothesis as to the causation of error , namely , that errors in rerum naturd result from the superposition of a large number of minuter errors arising from a , number of , independent sources .
The laws of these elementary errors are supposed entirely unknown , no further restriction whatever being imposed on the generality of the investigation ; as would be the case , for instance , were we to assume ( as has sometimes been done ) that each independent source gives positive and negative errors with equal facility .
To decide fully how far the above hypothesis ( which seems now to be generally accepted ) really agrees with facts , is an extremely subtle question in philosophy , -one which probably never can be more than partially resolved .
Still , even a cursory and superficial examination of a few particular cases seems to show that , far from being a mere arbitrary assumption , it is at least a reasonable and probable account of what really does take place in nature , in many large classes of errors of observations .
The history of practical astronomy , in particular , seems to prove that , whatever doubt may be entertained of its exactness as applied to the errors of rude and primitive observers , we may safely accept it in the case of the refined and delicate observations of modern astronomers .
It would be scarcely possible in this Abstract to convey any clear idea of the mathematical analysis employed in reducing the above hypothesis to calculation .
It will suffice to remark that , whereas in the processes given by Laplace and Poisson , when applied to the problem before us , the elementary component errors are at first supposed of finite magnitude , and finite in number , and the results are afterwards modified for the supposition that the magnitude of the errors becomes infinitesimal and their number infinite ; much simplicity is gained in this Paper by making these suppositions at the commencement .
Also , instead of taking a simultaneous view of all the elementary errors , as affecting the actual or resultant error , the latter is considered as produced by the superposition of some one of the elementary errors upon the error produced by the combination of all the others .
We are thus led to examine the infinitesimal change produced in a given finite error , as expressed by a given function , by the superposition of a new infinitesimal error ; and from the analytical expression arrived at , it is shown how to find the form of the function of error resulting from the combination of an infinite number of given infinitesimal errors .
This form is found to be altogether independent of the nature or laws of the component errors .
If we assume the following data as known , viz. m=sum of the mean values of the component errors , h=sum of the mean values of the squares of component errors , i=sum of squares of the mean values of component errors , it is proved that the probability of the actual resulting error being found to lie between x and x dv is 1 ( ' m)2 2(/ -i ) d : .
27r ( h-i ) This result will be found to agree with Poisson 's .

