XV .
" On the Theory of Probabilities .
" By GEORGE BOOLE , Esq. , F.R.S. Received May 21 , 1862 .
( Abstract .
) This paper has for its object the investigation of the general analytical conditions of a method for the solution of questions in the Theory of Probabilities , which was published in a work entitled " An Investigation of the Laws of Thought , &amp;c. " * The application of the method to particular problems has been illustrated in the work referred to , and more fully in a Memoir published in the Transactions of the Royal Society of Edinburgh , entitled " On the Application of the Theory of Probabilities to the Question of the Combination of Testimonies or Judgments .
" The latter contains also the foundations of the general analytical theory of the method .
But the complete development of that theory is attended with mathematical difficulties which I have only just succeeded in overcoming .
A correspondence on the subject between Mr. Cayley and myself also appears in the current ( May ) Number of the 'Philosophical Magazine , ' and I owe it to AMr .
Cayley that these further researches , of the results of which an account will here be given , were undertaken .
I shall make but few remarks here upon the i priori grounds of the method .
Generally it may be said that the solution of a question in the Theory of Probabilities depends upon the possibility of mentally constructing the problem from hypotheses which appear , whether as a consequence of our knowledge or as a consequence of our ignorance , to be simple and independent .
When the data are the probabilities of simple events , and no conditions are added , the problem is in theory sufficiently easy , the sole difficulty consisti ; ng in the calculation of complex combinations .
But when the data are the probabilities of compound events , or when the events are connected by absolute conditions expressible in logical propositions , or when both these circumstances are present , the difficulty of the required mental construction becomes greater .
If we assume the independence of the simple events from which the compound events according to their expression in language are formed , we meet , first , the difficulty that the number of equations thus formed may be greater or less than that which is requisite to obtain a solution ; and , secondly , the far more fundamental difficulty that the conditions under which the solution , supposing it to be obtained , is analytically valid may not coincide with the conditions under which the data are possible .
It seems indeed likely-at any rate the evidence of particular examples points uniformly to the conclusion-that any attempts to construct the problem upon hypotheses which , while not involved in the actual data are of the same nature as those data ( i. e. which might conceivably have resulted as facts of observation from the same experience from which the data were derived ) , limit the problem , and lead to solutions which are analytically valid under conditions narrower than those under which the data are possible .
But the processes of mathematical logic enable us , without any addition to the actual data , to effect the required construction of the problem formally-formally because the hypotheses which are regarded as ultimate and independent in that construction refer to an ideal state of things .
The nature of the conceptions employed , and their connexion with the conceptions involved in the actual statement of the problem , are discussed in the paper .
It is sufficient to say here that , whatever difficulty there may be in these conceptions as conceptions , there is nothing arbitrary in the formal procedure of thought with which they are connected .
The probabilities of the ideal events enter as auxiliary quantities into the process of solution , and disappear by elimination from the final result , but they are throughout treated as probabilities , and combined according to the laws of probabilities .
I will only say here that the difficulty which has been felt in the conception of the ideal events appears to me to arise from a misdirected attempt to conceive those events by means of the events in the statement of the problem the true order of thought being that the events in the statement of the problem are , not indeed in their material character , but as subjects of probability and of relations affecting probability , to be conceived by means of the ideal events .
Now the probabilities which constitute the actual data will in general be subject to conditions in order that they may be derived from actual experience .
Those conditions admit of mathematical expression .
Generally , if the events in the data are all or any of them compound , and if p , p , ... Pn represent their probabilities , those quantities will be subject to certain conditions , expressible in the form of linear equations or inequations , beside the condition that , as representing probabilities , they must be positive proper fractions .
All such conditions of either kind are ultimately expressible in the general form 61Pb + 6b22.+ bnPn+ AT the coefficients , , b , ... 6b , 6 differing in the different conditions so as to indicate that each of the quantities p , p2 , ... p , varies between a system of inferior limits expressed by linear functions of the other quantities , and a system of superior limits also so expressed .
Thus , if A , B , C represent any simple events , and if p represent the probability of the concurrence of B and C , p2 that of the concurrence of C and A , p3 that of the concurrence of A and B , then p , P2 , p3 must , in order that they may be derived from experience , satisfy the conditions P1 o ?
P2+3 3-1 , P2 J &gt; P 3+p1-l , 3 &gt; 1+)2 -1as well as the conditions implied in their being positive proper fractions .
On the other hand , the ideal events being by hypothesis simple and independent , the auxiliary quantities which represent their probabilities will be subject to no other condition a priori than that of being positive proper fractions-to no other condition a priori , because their actual values are determined in the process of solution .
Now the most general results of the analytical investigation are 1st .
That the auxiliary quantities representing the probabilities of the ideal events admit of determination as positive proper fractions , and , further , of a single definite determination as such , precisely when the original data supply the conditions of a possible experience .
2ndly .
That as a consequence of this the probability sought will always lie within such limits as it would have had if determined by actual observation from the same experience as the data .
The proof of these propositions rests upon certain general theorems relating to the solution of a class of simultaneous algebraic equations , and , auxiliary to this , to the properties of a functional determinant .
The following are the principal of those theorems:-1st .
If the elements of any symmetrical determinant are all of them linear homogeneous functions of certain quantities al , a , ... are -if the coefficients of these quantities in the terms on the principal diagonal of the determinant are all positive-and if , lastly , the coefficients of any of these quantities in any row of elements are proportional to the corresponding coefficients of the same quantity in any other row , then the determinant developed as a rational and integral function of the quantities a a2 , c ... a , will consist wholly of positive terms .
A nd , as a deduction from the above , 18 2ndly .
If V be a rational and entire function of any quantities x1 x2 ... xQ , involving , however , no powers of those quantities , and all the coefficients being constant , and if in general V , represent the sum of those terms in V which contain xi as a factor , and Vij the sum of those terms in V which contain the product xi xj , then the determinant w * ... v V1 V11 V12 V ... V '2 V21 '22 *** '2n W V V2 * ... V2.n will on development consist wholly of positive terms .
3rdly .
The definitions being as above , and the function V being in form complete , i. e. containing all the terms which by definition it can contain , the system of simultaneous equations VVVy V=PI --22 , * * grounds of the method , it is a perfect method of interpolation .
The analytical investigation , however , shows that , for the mere purpose of interpolation , the process might be modified by altering the coefficients of V without affecting its form ; but it indicates at the same time that such modifications have no definite analogy with that process by which weight is assigned to astronomical observations , and , from their arbitrary character , lead to results which cannot properly be regarded as expressions of probability in any sense .

