IL " An Inquiry into Newton 's Rule for the Discovery of Imnaginary Roots .
" By J. J. SYLVESTER , F.R.S. , Corresponident of the Institute of France .
Received April 6 , 1864 .
( Abstract .
) In the I Arithmetica Universalis , ' in the chapter " Id Resolutione Equationum , " Newton has laid down a rule , admlirable for its simplicity and generality , for the discovery of imaginary roots in algebraical equations , and for assigning an inferior limit to their number .
He has given no clue towards the ascertainment of the grounds upon which this rule is based , and has stated it in such terms as to leave it quite an open question whether or not he had obtained a demonstration of it .
Maclaurin , Camnpbell , and others have made attempts at supplying a demonstration , but their efforts , so far as regards the more important part of the rule , that namely by which the limit to the number of imaginary roots is fixed , have completely failed in their object .
Thus hitherto any opinion as to the truth of the rule rests on the purely empirical ground of its being found to lead to correct results in particular arithmetical instances .
Persuaded of the insufficiency of such a mode of verification , the author has applied himself to obtaining a rigorous demonstration of the rule for equations of specified degrees .
For the second degree no demnonstration is necessary .
For cubic equations a proof is found without difficulty .
For biquadratic equations the author proceeds as follows .
HIe supposes the equation to be expressed homogeneously in x , y , and then , instituting a series of infinitesimal linear transformations obtained by writing xa + hy for x , or y+ hx for y , where h is an infinitesimal quantity , shows that the truth of Newton 's rule for this case depends on its being capable of being shown that the discriminant of the function ( 1 , + e , e2 + e , l jx , y)4 is necessarily positive for all values of e greater than unity , which is easily proved .
HIe then proceeds to consider the case of equations of the 5th degree , and , following a similar process , arrives at the conclusion that the truth of the rule depends on its beintg capable of being shown that the discriminant , say ( D ) of the function ( 1 , ) C ) 2 , 7,2 ) 1n x. , y ) ?
, which for facility of reference may be termed " the ( e , ri ) function , " is necessarily positive when 04-e42 and i no are both positive .
This discriminant is of the 12th degree in E , ? .
But on writing x=ec , y=e 5+ i5 , it becomes a rational integral function of the 6th degree in x , and of the second degree in y , and such that , on making D=0 , the equation represents a sextic curve , of which x , y are the abscissa and ordinate , which will consist of a single close .
It is then easily demonstrated that all values of c , n which cause the variable point x , y to lie inside this curve , will cause D to be negative ( in which case the fanction c , 7 has only two imaginary factors ) , and that such values as cause the variable point to lie outside the curve , will make D positive , in which case the e ; n function has fouir imaginiary factors .
When the coniditions concerning c , i7 above stated are verified , it is proved that the variable point must be exterior to the curve , and thus the theorem is demonlstrated for equations of the 5th degree .
The question here naturally arises as to the significance of the sign of ID when such a position is assigned to the variable point as gives rise to imaginary values of e , , which in such case will be conjiigate quantities of the form X+ i , , X -ip respectively .
The curve D will be divided by another sextic curve into two portions , for one of which the couple e , ?
j corresponding to any point in its interior is real , and for the other conjugate .
This brings to viewt the necessity of there being in general a theory for equations with colnjugate coefficients , which for greater brevity may be termed conjugate equations , analogous to that for real equations in respect of the distinction between real and imaginary roots in the latter .
A conjugate equation is one in which the coefficients , reekoning from the two ends of the equation , go in pairs of the form p ?
iq , with the obvious condition that when there is a middle coefficient this must be real .
Such an equation may be supposed to be so prepared that , when thrown into the form P+iQ , P and Q shall have no common algebraical factor ; and when this is effected , it may easily be shown that the conjugate equation can neither have real roots nor roots paired together of the form X+4i , X-it respectively .
How , then , it may be asked , is the analogy previously referred to possible ?
On investigation it will be found that the roots divide themselves into two categories , each of exactly the same order of generality , -viz .
solitary roots of the form eiO , and as-sociated roots which go in pairs , the two roots of each pair being of the form 2i &amp; I0iO respectively ; so that , following the ordinary mode of geometrical representation of imaginary quantities , the roots of a conjugate equation may be denoted by points lying on the circumference of a circle to radius unity ( corresponding to solitary roots ) , and points ( corresponding to the associated roots ) lying in couples on different radii of the circle at reciprocal distances from the centre , each couple in fact constituting , according to Prof. W. Thomson 's definition , electrical images of each other in respect to the circle .
If the circle be taken with radius infinity instead of unity ( so as to become a straight line ) , then we have the geometrical eidolon of the roots of an ordinary equation , the solitary roots lying on a straight line , and the associated or paired ( imaginary ) roots on each side of , and at equal distances from the line .
In the inquiry before us , whether the variable point belong to the real or conjugate part of the plane of the D curve , it is shown to remain true that the number of associated roots will be two , if it lie inside the curve , and four if it lie outside .
The author then suggests a probable extension of Newton 's rule to conjugate equations of any degree .
In conclusion , he deals with a question in close connexion with , and arising out of the investigation of this rule , relating to equations of the form I+ ( ax+ 6)m = 0 , to which , for convenience , he gives the provisional name of " superlinear equations " ( denoting the ftnction equated to zero as a superlinear form ) , and establishes a rule for limiting the number of real roots which they can contain , which is , that if such equation be thrown under the form Xl(x+cl ) +Xs ( +C , )m+ ... .
and cl , c2 , .
e. cn be an ascending or descending order of magnitudes , the equation cannot have more real roots than there were variations of sign in the sequence X , , X2* X. , ( - ) MX1 This theoremn was published by the author , but without proof , in the 'Comptes Renidus ' for the month of March in this year .
But the method of demonstration now supplied is deserving of particular attention in itself ; for it brings to light a new order of purely tactical considerations , and establishes a previously unsuspected kind of , so to say , algebraical polarity .
The proof essentially depends upon the character of every superlinear form being associated with , and capable of definition by means of a pencil of rays , which may be called the type pencil , subject to a species of circulation of a different nature according as the degree of the form is even or odd , which he describes by the terms " per-rotatory " in the one case , and " 1 trans-rotatory " in the other ; so that the types themselves may be conveniienltly distinguished by the names " per-rotatory " and " transrotatory .
" Bv per-rotatory circulation i 's to be understood that species in which , commencing with any element of the type , passage is made from it to the next , from that to the one following , from the last but one to the last , from the last to the first , and so on , until the final passage is to the element commenced with from the one immediately preceding .
By trans-rotatory circulation , on the other hand , is understood that species in which , commencing with any element and proceeding in the same manner as before to the end element , passage is made from that , not to the end element itself , but to its polar opposite , from that to the polar opposite of the next , and so on , until the final passage is made to the polar opposite of the element commenced with , from the polar opposite of its immediate antecedent .
The inumber of chaniges of sign in effecting such passagoes , whether in a per-rotatory or a trans-rotatory type , is independent of the place of the element with which the circulation is made to commence , and may be termed the variation-index of the type , which is always an even number for per-rotatory , and an odd number for trans-rotatory types .
A theorem is given whereby a relation is established betwee n the variation-index of a per-rotatory or trans-rotatoryand that of a certain trans-rotatory or per-rotatory type capable of being derived froil them respectively ; and this purely tactical theoremn , combined with the al , ,ebraical one , that the formf(x , y ) cannot have fewer imaginary ftctors than any linear combinlation of df q ?
, L-ads by successive steps of induction to the theorem in question , dx die but under a more general form , which serves to show intuitively that the limit to the number of real roots of a superlinear equationl which the theorem furnishes must be independent of any homographic transformation operated upon the form .
The author believes that , whilst it is highly desirable that a simple and general method should be discovered for the proof of NYwton 's rule as applicable to equations of any degree , and that the strenuous eforts of the cultivators of the New Algebra should be directed to the attainment of this object , Ihis labours in establishing a proof applicable as fair as equations of the 5th degree inclusive will not have been uniproductive of good , as well on account of the confirmation they afford of the truth of the rule , towards the establishmenit of which on scientific grounds they constitute the first serious step yet made , as also , and still more , by reason of the accessions to the existing field of algebraical speculation to which they have incidentally led .

