" On the Mysteries of Numbers alluded to by Fermat .
" By the Rt. Hon. Sir FREDERICK POLLOCK , Lord Chief Baron , F.R.S. , &amp;c. * ( Abstract .
) This paper is presented as a continuation of one which appeared in the Philosophical Transactions of the Royal Society for 1861 , vol. cli .
p. 409 , and the object of it is to call attention to certain properties of odd numbers when placed in a square , according to an arrangement to be explained below .
It appears to me probable that these properties are connected with ( if , indeed , they be not actually some form of ) the mysterious properties of numbers , to which Fermat alludes in the announcement of his theorem ( as furnishing the proof of it ) ; for in point of fact these properties give a method by which every odd number can be divided into four square numbers , and every number ( odd or even ) can be divided into not exceeding three triangular numbers .
I am not prepared to say whether or not the method affords a demonstration , and proves that it can always be done ; but it always does it , and the cause of its success may be distinctly shown .
The properties I allude to are scarcely less interesting and curious than the theorem itself , and present results for which I can find no namne more appropriate than the geometry of numbers , for relations appear to be established between various numbers in the square , which relations are not founded on any arithmetical connexion between them , but on the positions they respectively occupy in tne square of which they form a part .
The arrangement of the numbers is as shown in diagram No. 1 .
Any odd number ( which may be the subject of inquiry ) is made the first term of a series , increasing from left to right by the numbers 4 , 8 , 12 , 16 , ... ( 4 n ) This series forms the horizontal line at the top ; each term of the series so formed is made the first term of a series , increasing downwards by the numbers 2 , 6 , 10 , 14 ... .
( 4n-2 ) .
A square of indefinite magnitude is thus formed , consisting of two sets of series , one set all horizontal , the other set all vertical .
161 is the first number in diagram No. 1 .
The result of the arrangement of the two series in the manner above mentioned is the formation of a third set of series , which may be found in the diagonal lines of the square .
If from the first number in the square ( 3 a diagonal linie be drawn towards the opposite corner , it will pass ugh the first terms of one portion of the third set of series ; the second , third , and following , terms are takeni alternately from each side of the diagonal line .
These series increase by 2 , 4 , 6 , 8 , &amp;c. ... ( 2n ) , and with the exception of one term they are all in the diagonlal lines ( see diagram No. 1 ) , in which the single red line passes through the first terms , and the double red line shows where the terms of the series ( the second , third , &amp;c. ) belonginog to that ( as a first term ) are to be found ; the red inik numbers indicate their order .
12345678 The Nos. 203 , 205 , 209 , 215 , 223 , 233 , 245 , 259 , &amp;c. , &amp;c. , compose the series ; the terms increase by 2 , 4 , 6 , 8 , 10 , &amp;c. ... 2n .
Any number in the diagonal from 161 may be the first terni of a similar series .
The diagonal from 163 will give all the other first terms of the third set of series .
In order to explain the indices which appear in the diagram No. 1 , and to show in what manrner the series are connected with , and pass into each other , it is necessary to point out the properties of the two series which compose the square , and of the third series , which is a niecessary result .
All of them have this property in common , -that if you can discover the roots of the squares which compose any term of the series with reference to the nature of the series , and the order in the series of that term , then you know the roots of every term in the series , both before and after that term .
The first series iniereases by 4 , 8 , 12 , &amp;c. The indices of the terms of a series so increasing I have made 1 , 3 , 5 , 7 , &amp;c. , and they are put at the top as being comimoni to all the series that are horizontal .
For this reason , if two numbers differ by 1 , as n , nz+ 1 , and the larger be iniereased by 1 , and the smaller be diminiished by 1 , and the process be continued , the result will be .
21 , n+ 1 , n-1 , ?
I +-2 , nz-2 , n+3 , n-3 , n ?
+4 , &amp;c. &amp;c. n--(p-1 ) n+p .
If these be treated as roots , the sums of the squares will be 2 2+ 2 + ?
I. 2n'+22z+5 , 2n2+2n+ 13 , 2n2 ?
+ 2n + 25 , &amp;c. &amp;c. 2n2+2n+ 2P2_2p + 1 .
The sums of the squares inierease by 4 , 8 , 12 , &amp;c. If therefore the roots of the square , into which any odd number may be divided , bep , q , n , n+ 1 , and the number be increased by 4 , 8 , 12 , &amp;c. ... ( 4n ) , the mnth term in the series will be composed of squares whose roots will be p , q , n-(n-1 ) , n + m ; two of the roots will be constant , the others will vary , and their differences in the successive terms will be 1 , 3 , 5 , 7 , &amp;c. ( 2n1 ) ; and if you discover the roots of any term in the series , you can find the roots of all the terms .
The second series resembles the first in having two roots constant , and two variable ; the differences between the variable roots are , in the first term 0 , in the second term 2 , in the third term 4 , &amp;c. , and the indices of the terms are therefore 0 , 2 , 4 , 6 , 8 , 10 , &amp;c. , which are placed vertically by the side of the square .
For if two numbers are equal , as ni , An , and one of them be inclreased by 1 , and the other diminished by 1 , and the process be continued , the result will be nn 2z n-1 , n+1 j If these be treated as 2N2 ?
+2 n-2 , n+2 &gt; roots , the sunms of &lt; 2n'+S n-3 , n+3 their squares will be 2n 'd 18 &amp;c. &amp; C. J &amp; C. &amp;c. The sums of the squares increase by 2 , 6 , 10 , 14 ... .
( 4nz-2 ) , and the successive differences of the variable roots are 0 , 2 , 4 , 6 , &amp;c. ; and if the roots of the squares inito which any odd number may be divided be p , q , n , z , and the number be increased by 2 , 6 , 10 , &amp;c. , the roots of the mth term will be p , q , n-(mr-l ) , n ?
+(n-l ) , and if the roots of any one term be known , the roots of all the others may be found .
The small figures in the upper right-hand corner of each division or small square are the indices of the third set of series .
In this set all the roots are variable .
The character of the first set of series is , that two roots in every term differ by an odd number ; the character of the second set of series is , that two roots in every term differ by anl even number ; but in the third set of series , the algebraic sum of all the roots of the squares into which the successive terms may be divided is successively 1 , 3 , 5 , 7 , 9 , &amp;c. ( an odd number ) : the sum of the roots of the squares into which an odd number can be divided cannot be an even number .
The following Table will explain in what manlner the series is formed from the ropts of the squares into which any odd number may be divided , so as to make the algebraic sum equal to 1 .
I have preferred to use figures instead of algebraic symbols , as being more readily understood and more easily dealt with ; but the result is the same whatever figures or symbols may be used .
The series begins from the centre .
Let-7 , -3 , 2 , 9 , which are the roots of the squares into which the odd number 143 may be divided , be placed in the centre , and let the positive roots be increased downwards and decreased upwards , and the negative roots increased upwards and decreased downwards , the result will be as in the Table below : Order of Algebraic Sums Order of Roots .
sums of of squares of terms .
roots .
of roots .
terms .
457 12 -13-9-4 3 -23 275 12 10 -12-8-3 4 -19 233 10 8 -11-7-2 5 -15 199 86 -10-6-1 6 -11 173 64 9-5 077 155 42 8-4 183 145 2 Centre 1 7-3 291 143 13 6-2 3710 5 149 3 5-1 4 11 9 163 574 05 12 13 185 79316 13 17 215 911227 14 21 253 11 13 1-3 8 15 25 299 13 &amp;c. &amp;c. &amp;c. It will be observed that the terms of the series 143 , 145 , 149 , 155 , &amp;c. increase by 2 , 4 , 6 , 8,. .
( 2 n ) .
In the column of the sums of the roots 1 , 5 , 9 , 13 , &amp;c. increase by 4 .
1 , -3 , -7 , lII decrease by 4 ; the differences of the roots , if arranged in the order of their numerical value , is always the same throughout the series .
Note .
In the remainder of this paper every odd number that becomes a term in any of the series is expressed by the roots the sum of whose squares form the number itself ; 2 , 3 , 6 , 8 means that the number occupying that division or small square is 113= ( 22+32+ 62+8 ) ; a figure ( or collection of figures ) representing , merely the arithmetical value is put into it circle , thus : 2 , 3 , 6 , 8 Having explained the construction of the square and the indices which belong to the series of which it is composed , I propose to point out the properties which discover the roots of the squares into which the odd numbers ( which are found in the different parts of the square ) may be divided .
If the first odd number in the square be of the form ( 4 p+ 2 ) .
n+1 ( or 2 n+ 1 , 6 n+1 , 10 n+l , 14 n+1 , &amp;c. ) , n being of any value whatever , the ( n-p)th term will be ( p+ l ) , -p , n , n j ; these are the roots ( the number itself would be 2 n2 ? ?
2p2 + 2p + 1 ) ; and as the index of the nth term is n+(n-1 ) , or 2n-1 , the inidex of the ( n-p)th will be 2 n-(2 p+ 1 ) , and the algebraic sum of the roots may be made equal to the index ; it is therefore a term in a diagonal series Lthat the ( n-p)th term is 2 n+ 2 2p2+ 2p+ 1 , will appear by finding in the usual way the ( n -p)th term of a series whose first term is ( 4 p+ 2 ) .
n+ 1 , and the terms of which increase by 4 , 8 , 12 , 16 , &amp;c. ... 4 n ] ; but as two of the roots are cqual , n , n , it is the first term of a vertical series descending , thus : p+ l , p , n , n p++ p , ( n-1 ) , n+1 p+I , p , ( n-2 ) , n+2 p ?
+ p , , ( n-3 ) , ( n+3 ) .
I call this term ( A)* and indicate it by that letter , and from this term the roots of many others may be derived ( which are indicated by other letters connected with A in an invariable manner ) , whose squares will compose the number that belongs to that term .
For example , counting 2p ? ?
1 squiares backwards from A is a term I have distinguished as W ; the roots of the number belonging to it are p +1 , ) , ( n)-4p+2 ) , n These roots may , of course , be either positive or negative ; arranging them thus , -p-+ 1 , -p , 4p+ 2 , n , we have the algebraic sum of their roots equal to 2 n6 p+ 3 , which is the index of the square in which W is found going backwards from A , the index of which is 2 n2p +1 ( as already stated ) ; immediately adjoining W are two squares which I have called M and N respectively .
M is the term next before W , and is composed of the roots n-2p ? ? ?
?+2 , it-2 p+2 , 3p+2 , p+ 1N is the term next after W , and consists of the roots n-21p , n-2p , 3p+l , p1 Each of these will traverse the squiare diagonally , the algebraic sums of their roots beirig equal to the index ; the roots of W may also be obtained from A by takin-g its roots dowln vertically , making n , n , successively n1 , n+1 , n-2 , + 2 , &amp;c. , until the square is reached , in which the index is 2 n+ 2p+ 1 , and 2 i+ 2p+1 being the arithmetical sum of all the roots of A ; A here becomes a term in a series which moves diagonally , and on being carried up towards the left will give the roots of W. Whclu the indices of the squares through which this vertical series passes equal 2n+ 1 , by makingjp + 1 , p , one positive and the other negative , the term becomes a term in another series which moves diagonally upwards towards the left .
This must occur both when the index equals 2 tI and when it equals 2n +1 ; and as the indices increase downwards uniformly by 2 , it follows that 2 ni-1 and 2 n+ 1 will be the indices of contiguous terms of the vertical series , and therefore two contiguous terms will become terms of series moving diagonally upwards to the left ; and as these two series are contiguous to each other , their terms found in the first series ( that is , the series in the top line ) will also be contiguous .
These two terms I have designated as AM and AN .
AM comes from the term where the index equals 2 n+ 1 , and AN from the terii where the index equals 2 n1 ; the roots of AAI are 0 , 2p+1 , 2-1)n2 , those of AN are 0 , 2 +l , m-2p , I ' andi being arranged thus , --2 )+ 1 , 0 , m-2p+ 2 , it , -2 p+ 1 , 0 , m-2p2 , m , will move diagonally downwards to the left ; and as each of these have two r'oots that differ in onae case by 2 I , in the other by 21+ 2 , the terms in these series that arc parallel to those terms in the first vertical series which have their externlal indices respectively 2 1 ) and 2p+ 2 , the terms of A:iV and AN will ( I say ) in these places become terms in series movirig vertically ; and on being followed up to the series in the top row will be found to give the roots of TMI and N. The roots of MA and N may be -obtained by another method as follows:-As the algebraic sum of the roots of A equals its index , therefoie A is a term in a diagonal series moving downwards towards the left ; and as two of its roots , p ) , i ) + 1 , differ by 1 , it follows that whenever the value of m has been so altered that 2n+1 equals the inidex by making + 1 , p , one positive and the other negative , the term becomes a term in another series , which series will move at right angles to the series last mentioned .
Now taking the series which moves to the left of A , it is clear that this result will obtain in .
two places ; first , whern the altered value of i makes '2 -1 equal the index , and secondlv , when the altered value of n makes 2 n+1 equal the index .
The first of these going up gives the roots of N , the second gives those of M , and these roots are identical with those obtained from AM and AN .
The index of A is 2 i2 p+ 1 , therefore when n in moving downwards becomes n+1 , n+2 , + ?
3 ... ... . .
2nz-3 p-F2.2 n-3_p+l + 1 , p being two of the roots of each term in this series ; by adding , p ? ?
1 to the first-mentioned root and p to the other , these two terms will be found to be terms in two horizontal series , of which the first terms are in the first vertical series , and these terms both of them make the diagonal index , and therefore are terms in a diagonal series which , rising towards the right , give the roots of W. A descends diagonally to the left , and on each side of the line which leads to W changes to cross diagonals which lead to M and N. W leads diagonally to the left to R and S , and where it crosses AT changes and goes up to A. M goes down into R , and then diagonally to where it meets the horizontal series from T ; its roots there correspond with the series from T ; it returns to T and uip to A. In like manner N goes down through S to the line from V , and so to V and up to A. R and S go each of them across to the vertical line from A , and so up to A : every term through which these lines pass has the four roots indicated whose squares would make the term .
The number of terms in the whole square , whose four roots may be expressed in term of p and n , is very considerable ; and it may be well now to present some skeleton diagrams of the many ways in which certain members of the square are in7varia.6y connected .
I propose to exhibit several ( to avoid confusion , which would arise from putting all in one diagram ) ; these do not by any means include all the connexions that exist ; but whatever may be the value of n or p , a number of the form ( 4p+2).n+1 , whether it be 2n+1 , 6n+l , OnX+1 , &amp;c. , and whatever be the value of et , gives the following resuilts .
See diagram No. 2 .
At the ( n-p)th term there will be A , which descends vertically till the index is 2n1 , then 2n+ 1 , and from these rise up in diagonals AN and AM , as already mentioned .
A then further desceinds till the index is equal to 2n + 2p + 1 , when it rises in a diagonal to W. Diagram No. 3 shows certain connexions between AM and AN and other terms in the square .
AMt is always -2p-+1 , 0 , n-2pp+2 , ?
n , AN is always -2p + 1 , 0 , n -2p , n. If the series in which AMN is a term be carried down diagonally till 0 becomes 2p +1 ( that is 2p +1 places ) , it becomes a term in a diagonal series that intesects it and rises to the top , then goes down to the horizontal series from It , where it becomes a term in that series and passes to where it is below M and rises vertically up to it .
AN does the same with respect to the series from S , and rises up to N. If the term AM descends till n 2p +2= 2p 1 , it rises to the left in another diag , onal series and goes on till it crosses M , where it is foiind that the roots are always the same as those which arise from M , descending by means of its two eqiual roots .
The term AN does the same with respect to N. If AMIf descend to the margin at am , and one step further into an , and AN descends to an , they will be found to have the same roots , and they will be _ from AMi _ -2p+l , 1 , n-2p+1 , n from AN -2p+ 1 , -l , n-2p+1 , n The only differenice being that in the one case 1 is positive , in the other it is negative ; but whatever be the value of p or n , in this portion or term of the square 1 is always one of the roots .
In crossing the two horizontal series from R and S , it will always be found that at the points of intersection the roots of AMi corresponed with the roots of the series from I , and the roots from AN correspond with those from S. Diagram No. 4 exhibits the way in which B , C , P , and Q are connected together .
The roots of A at the ( it-p)th square will always be -(+ 1 ) , -p , n , n , aidp ?
+ 1 , p must be one of them odd , the other even ; therefore , whether n be odd or even , ain odd number will be formed by -(p+ 1 ) , t , or -p , n. The series from A descending diagonally has the roots it , i decreasing , but the niegative roots iniereasing .
The differences will continue the same ; and when the series arrives under that index which corresponds to the odd number -(1-)+ 1 ) n , or --p , n , it becomes a term in a horizontal series which goes to P , two terms of which are always p , p+ 1 .
W , in descending diag , onally , has its index on reaching PB 1 .
When A has descenided so as to reach the evenl number of the two , ( -(p+1 ) , n , -_ ) , n ) , it rises in a vertical series to Q ; and the three series , WR .
, BP , CQ , always intersect in the same point or small square , H. The paper then exhibits in a Diagram ( No. 5 ) all the roots which arise from applyij g the method to the odd No. 161 , and shows that the roots of A would be ... ... . .
3 , 2 , 16 , 16 ofW ... ... ... ... . .
3 , 2 , 6 , 16 of Mt , ... ... ... ... .
8 , 3 , 10 , .
10 of N , , ... ... ... ... .
7 , 2 , 12 , 12 of AM , ,5 , 0 , 10 , 16 of AN , ,.5 , 0 , 12 , 16 It shows the roots of the squares marked B , C , P , Q , R , S , and many others ; and , finally , it showrs that 161 would be composed of the squares of the following roots , either 3 , 12 , 2 , 2 , or 10 , 0 , 5 , 6 ; but to set forth the roots of the numbers which are iu Diagram No. 1 , would require a diagramn larger thaui could conveniently be put into a publication of the size of the Proceedings .
The roots 10 , 0 , 5 , 6 , if arranged thus , -10 , 0 , 5 , 6 , have 1 for their sum ; but it was proved in the former paper ( see p. 410 , Trans. Roy .
Soc. for 1861 ) that if the algebraic sum of the roots be 1 , then the number is the double +1 of a number composed of 3 triangular numbers , 16 1=80 x 2+1 , and 80 is composedof the 3 triangular numbers 55 , 15 , 10 .
If therefore any number be doubled , and 1 be added , aln odd number will be obtained , to which the same process may be applied as is here applied to 161 .
I have stated that the cause of the success of " the method " ( though it does not at present amounit to a demolnstration ) may be easily slhown .
It arises , first , from " the mnethod " requiring every odd number that is a term in any of the series to be represented by the roots of the square numbers that compose it ; and secondly , and more particularly , from every series being connected with at least six others of a different kind which intersect it , each of which is again connected with at least five others , so that when the whole nietwork has been pursued , and the roots which in successioni form every term have beeni recorded , it will be fouild that many different modes of dividing each term into four squares or less will be discovered , i. e. if the numbers be large .
I propose to show the manner in which the series are apparently interwoveli by an example from each kind .
21 23 Let the first term in a horizontal series be 22 3 , 7 , 13 , 14 with 22 as the index in the margin , and 21 and 23 being , the indices of the diagonal series which pass through this square ; for , except at the top line , two diagonal series pass through every square .
13 and 14 are the variable roots , which become 12 , 15 , 1 11 , 16 , 1 10 , 17 , 1 9 , 18 , l &amp;c. t in the successive terms ; when in the second term 14 becomes 15 ( 15+7=22 ) , and the roots are 3 , 7 , 12 , 15 , a term in the series which would come down from the second square ; the roots in that square will therefore be 3 , 12 , 4 , 4 ; when 15 becomes 19 the roots will be 3 , 7 , 8 , 19 , and the roots at the top will be 7 , 8 , 8 , 8 ; so when 15 becomes 25 at the tenth term the roots are 3 , 7 , 2 , 25 ; and as 3 , 25=22 , another series rises up with roots of the first term , 7 , 2 , 14 , 14 .
13 diminishes to 0 , and then increases , giving 2 more when it becomes 15 or 19 ; but the series is also crossed by diagonal series ; and when the 21 19 17 iS 13 , &amp; cj.3 index in the two rows from the first term 23 25 27 29 31 , &amp;c. is37 ( the sum of all the roots ) , or 31 ( the sum of the variable plus the differenlce of the constant roots ) , or 17 ( the sum of the variable roots minus the sum of the constant roots ) , or 23 ( the sum of the constant roots minus the difference of the variable ) , a diagonial series arises .
Hlere are no less than 10 other series by which this is crossed and associated and connected , and the number cannot in any case be less than 6 .
A series of the second kind gives rise to other series crossing it in the same manner mutatis mnutandis , which result is so obvious that it is not necessary further to dwell upon it .
A series of the third kind has all the differences of its roots the 13 same in each term .
Let -7 , 2 , 4 , 14 be a term in a diagonal series at the top row of the system .
11 13 15 92 10 0 8 , 13 , 2 , 2 -7 , 2 , 4 , 14 6 , 15 , 4 , 4 Thus : __9 ~~~~~~~~~17 92 10 92 To 2 -8 , 1,3 , 13 -6 , 3 , 5 , 15 When it reaches to the left and to the right , the second place , as 3-1=2 and 5-3 2 , it furnishes two vertical series ; at the tenth row it furnishes two more ; at the twelfth row two more .
Wheni it gets into the column whose index is 11 and then 9 , before it reaches the margin or outer edge , it furnishes two horizontal series ; and after it has passed the margin at 9 and 11 it furnishes two more , and at 21 it furniishes another .
Here are six new vertical and five new horizontal series ; besides which it fiurnishes at least two other diagonal series which cross it .
Having stated the properties which belong to the square , if the first odd number in it be of the form ( 4p + 2).n ? ?
1 , and that whether it be 2n + 1 , 6n ? ?
1 , l On ? ?
1 , &amp;c. , or whatever be the value of ni , certain squares may be found which I distinguish as A , B , C , H , P , Q , W , M , N , AM , AN , R , S , T , V , &amp;c. , which are connected together by a community of roots where the series cross each other in a manner that is invariable .
The Diagram No. 3 is another example of the manner in which certain of the terms in the different series communicate with each other , by the roots being comMon to both , at the point where they cross .
AM passes diagonally to am , down to an , and up to AN .
AN , in like manner , goes down to an and up to am and AM .
If the first term of the square be an odd number of the form ( 4p+2 ) n+1 , the roots from AM are in an -(2p + 1 ) , 1 , ( n-2p + 1 ) , n. The roots from AN are ( 2p + ) -1 , ( n-2p+ 1 ) , n. The iindices of the diagonal series are 2n-(4p+ ) and the algebraic sum of the roots is the one or the other , according as 1 is + or- ; but AM also passes to the series from1 R , and AN to the series from S , and go to R and S , and thus go up to W. AM also reaches the vertical line from M , and passes up to M , as AN does to N. Lastly , AM goes to N thus , and AN to W in a similar manner .
Whatever be the Proc. Ror .
Soc. Vo0i XVPl .
Ill. D)iagram IN ?
Il I___ _3579I 13 15 17 19 10on + 131 n]=is 01 165 173 185 201 221 2 273 305 341 35317 187 923 27 15 ; 41 1697'3 193 2 229 253 281 349 64 179 183 199 2 219 239 263 291 323 9715 17 29 X2|1 191 273 913 25 , 871 93 1 197 [ } &lt; 1 217 2 253 277 305 337 9 373 691 55 17 I9 21 23 22 29 27 2 12 1 237 245 257 273 293 3+5 377 413 |41 | 23 1271 1283 2 99 3139 3 43 13 I 403 1439 161 94 25 301 313 29 349 1 373 401 3 469 1 19 21 253 25 2 29 31 33 35 37 18 21 231 227 2 &gt; 347 3563 381 407 435 467 503 2 21 23 25 7 27 9 31 33 35 37 39 201 361 365 2 373 5 401 421 445 473 505 541 values of p and n , these connexions occur , but the form of them varies with the values of p and n. I have pointed out some of the results of having as the first term in the square an odd number of the form ( 4p + 2 ) , n+ 1 ; the forms will be 2n + 1 , Gn+1 , lOn+1 , 14n+1 , &amp;c. , in which p may be 0 , 1 , 2 , 3 , &amp;c. But there is another form which gives similar results , viz. 4pn + ( 2p + 1 ) , which is in many respects the " converse " of the other ; the forms will be 4n + 3 , 8n+5 , 12n+7 , 16n+9 , &amp;c. These are the first terms to which this system gives rise .
The ( n-p-1 ) th term is always Pp , p , n , n+ 1 .
In the former system n produced the equal roots and p the uniequal ; here , p produces the equal and n the unequal roots , and 1 call the ( n-pI)th term A a &amp; .n the other .
This system has also W and M and N on each side of it , and other squares similar to the other , but the roots of which they are composed are differenitly formed .
M and N come from below .
W , instead of being -p+ 1 , -p , ,n-4p+2 , n , will be -3p , p , ( 'n-2p ) , ( i ? ?
1 -2p ) ) , and other terms are similarly altered , but the general result is the same .
A specimen of a square of this form is given in Diagram No. 8 , but which canniot be reduced to the size of the Proceedings , where p=3 andn=16 ; the first number is 199 , and the square is completed so far as to show 5 , 13 , 1 , 2 that the roots of the 4 squiares whose sum is 19)9 , -3 , l , 10 1 , 10 ) , 7 , 7 but neither in this Diagram nor in Diagram No. 5 is the whole square completed ( to avoid confusion ) ; but if the series be traced in succession , the entire Diagram 6 would be filled up , and every term would disclose the roots whose squares compose it .
In this mannier every odd number in all the series is divided into the squares that compose it ( not exceeding 4 ) , the squares being indicated by their roots .
The two systems of ( 4p ? ?
2 ) xn +1 and 4 1.n ?
+2p +1 include every possible odd numlber ; 4n +3 includes every alternate odd number from 3 ; 8n+5 every fourth number from 5 , and so on ; 2n ?
+1 includes every odd number ; Gn ? ?
1 every third odd number .
Many odd numbers belong to both systems , and to more than one in each .
15 1 is an examnple ; it is either 10(n +1 ( n-=15 ) , or it is 12n+ 7 ( n= 12 ) .
The paper contains a Diagram ( No. 9 ) exhibiting the odd number 151 ) I as belonging to both systems ; but the Diagramn cannot be redu-ced .
The roots of the squares that compose 151 are ( 10 .
1.5. .
5 ) , or ( 3 .
9 .
5 .
6 ) .
i+.z E n ?
Cs 0 F ; ~= 4wIo 0t12~~~~t cec~~~~~~~~~c 01_ 2 dF DX2X0 bnCD ? .
) H~~~ '1t ' Q 01 C. ) a cC.~~~~~~~~C n^tn Qq Fil t~~~~E cEI 4S : , a 01 _____ 0CI)S.~~~ ~~ 1 ?
.~ -~ 4 ) I I~~~~~~~~~~~~)ht 0 r01~~~~~~~4 + 01 .
011 I U2~~~~~ '~~~gt Diagram No. 7 .
There is an interval of n-3p +3 squares from the 1st term .
*2n6P+4 *2n-6p Jr 4 p-+2 , pn 21)+ 2 , nX2p+ I __ 2 6+ 1 This Diagram ( No. 7 ) 2n-6l ) +4 -39+ 12 shows in terms of n and p the SX+ 1 , P-2p roots of X , S , an ( which ___ 2_____ _2p _ comes down diagonally from AN ) , T , and V , and the There is here an interval of number of squares between ( p-I ) squares from S to an .
them ; the relative positions --______________ _ of these terms depend entirely 2n-4p +3 on p , and are always the + same for the same value of 2n 4.2 +2 _ , -p 2p P1 } an -_+_ , I Terms similar to these in n2p + 1 , n crease indefinitely as n increases .
The value of cerThere is here an interval of taimi roots is independent of ( p-1 ) squiares from an to T. i , and therefore is the same -____________________ __ w Ufor every value of n. 2n-2p32 T p11 + I , p+l n-1 , 2n 2n-2p 2n-2p-1 v P,.r %pn , n+ 1

