BOSIDNCOUEGESCiENCEUBR/iri? £rf f2. Digitized by the Internet Archive in 2010 with funding from Boston Library Consortium IVIember Libraries http://www.archive.org/details/treatiseongeometOObass A TREATISE ON THE GEOMETRY OF SURFACES i,^.^ ^^U^i^^ ON THE GEOMETRY OF SUEFACES BY A. B. BASSET MA. F.R.S. TRINITY COLLEGE CAMBRIDGE MATH. DEPT. BOSTON COLLEGE LIBRARY CHESTNUT HILL, MASS, CAMBRIDGE DEIGHTON BELL AND CO. LONDON GEORGE BELL AND SONS I910 [AH Rights reserved'] (Cambritige : PRINTED BY JOHN CLAY, M.A. AT THE UNIVEKSITY PRESS. 1937DE PEEFACE THE last edition of Salmon's Analytic Geometry of Three Dimensions, which was published in 1884, has been out of print for some years ; and although there are several excellent works on Quadric Surfaces and other special branches of the subject, such as those of Mr Blythe on Cubic Surfaces and of the late Mr Hudson on Kummers Quartic Surface, yet there is no British treatise exclusively devoted to the theory of surfaces of higher degree than the second. I have therefore endeavoured to supply this want in the present work. The Theory of Surfaces is an extensive one, and a thoroughly comprehensive treatise would necessarily be voluminous. I have therefore decided to limit this work to the more elementary />^ a i^^ portions of the subject, and have abstained fronTrTntroducing ^>^^/ inves¥igations^"wliich require a knowledge of the Theory of Cm^^ Functions and of the higher branches of Modern Algebra. The|^^j^» — 6 on the Hessian 290 IV. Quartic scrolls of the 13th species 291 ERRATA Page 2, last line, add or constant multiples of the distances. „ 4, line 7, read points for point. „ 5, equation (6), delete ... after second term. „ 6, line 15, read -tion for -tion. „ 49, line 6 from bottom, read 2va8 for 2av8. „ 67, line 10 from bottom, read is not a prime for is a prime. „ 74, line 9, read dUjd^ for dFjd^. ,, 81, heading, read Trisecant for Trisectant. ,, 136, line 13 from bottom, read k for 2k'. „ 164, line 8, read lm% for three. „ 168, line 10 from bottom, read conic node for conic. „ 175, Sections 239 and 240 require correction. Weddle's surface is the Jacobian of four quadrics. The surface (6) is a particular case of Weddle's surface. „ 177, heading, read Octonodal Quartics. „ 205, line 19, read a{pa^+fky^+gk(iy) + '2,Rky^ for (a, /3, yf. „ 233, line 7 from bottom, read S{\, 1, n) for S{\, 1, n). CHAPTER I THEORY OF SURFACES 1. The general equation of a surface of the nth. degree, when expressed in Cartesian coordinates, is tlo + Ui + U2+ ...ltn = (1), where Un is a ternary quantic of {x, y, z). The number of terms in (1) is equal to the sum of the series i{1.2 + 2.3+...0i + l)(w + 2)}, that is to say i (n + 1) {n + 2) {n + 3). The number of independent constants in (1) is one less than the preceding quantity and is therefore equal to ^n{n^ + Qn + 11), which determines the number of independent conditions that a surface of the nih. degree can satisfy. If in (1) we put y = z = 0, we obtain an equation of the nth. degree for determining the points where the axis of a; cuts the surface. Hence : — every straight line intersects a surface of the nth degree in n points. Also if we put z = in (1), we obtain an equation of the nth degree in x and y, which determines the curve of intersection of the plane z = Q and the surface. Hence : — every plane intersects a surface of the nth degree in a curve of the same degree. Let Un = 0, Vm, = be two surfaces of the nth and wth degrees respectively ; then if z be eliminated we shall obtain an equation of degree mn in x and y, which represents a curve of degree mn which is the projection on the plane ^ = of the curve of intersection of the two surfaces. Hence : — two surfaces of degrees m and n intersect in a curve of degree mn ; also every plane intersects the curve in mn points. B. 1 2 THEORY OF SURFACES Three surfaces of degrees I, m and n intersect one another in linn points ; for if Ui = 0, F^ = 0, Wn = be the three surfaces, it is shown in treatises on Algebra that the result of eliminating y and z is an equation of degree Imn in x, which determines the values of x at the points of intersection of the three surfaces. And by parity of reasoning it follows that : — every curve of degree n intersects a surface of degree ni in mn points. The curve of intersection of two surfaces does not in general lie in a plane. There are consequently two kinds of curves in space, called plane and twisted according as they do or do not lie in a plane. It may also happen that a curve which is apparently a twisted one may degrade into two plane curves lying in different planes. 2. There are four distinct species of surfaces. First ordinary surfaces such as (1). Secondly scrolls, which are also called skew surfaces. These are generated by the motion of a straight line which moves in such a manner that two consecutive generators do not intersect. Thirdly, when each generator intersects the consecutive one the surface is called a developable* surface, because it is capable of being unrolled into a plane. Fourthly cones, which are a special kind of developable surface, in which all the generators pass through a point. A cylinder is a special kind of cone which is obtained by projecting the vertex of the latter to infinity. It will hereafter be shown that all surfaces of a higher degree than the third are capable of assuming all four forms, provided they possess certain singularities. Thus a quartic surface which has a triple line is a scroll ; if it has a cuspidal twisted cubic curve, it becomes a developable surface ; whilst if it has a quadruple point, it becomes a cone. Quadriplanar Coordinates. 3. In the quadriplanar system of coordinates, the position of a point P is determined by its distances (a, /3, y, S) from the four * Some writers call a developable surface a torse. This is an inaccurate use of language, because torse is derived from torsi the perfect of torquere, to twist; and since a developable surface is formed by bending a plane, and therefore involves the idea of flexion alone, a word which connotes torsion is altogether inappropriate. The Italians call a developable surface una sviluppahile , and a scroll una gohha. Both scrolls and developables are included in the general term supcrjicic rignti' or ruled surfaces. QUADRIPLANAR COORDINATES 3 faces of a tetrahedron ABGD, called the tetrahedron of reference. The coordinate a is the length of the perpendicular drawn from P to the face BCD, and is positive or negative according as P lies on the same or the opposite side as ^. If a, h, c, d be the areas of the four faces which are respectively opposite to A, B, G and D, and V the volume of the tetrahedron, then which we shall write 1=1 (2) ; accordingly any algebraic function of the coordinates, which is not homogeneous, may be made so by multiplying each term by the proper power of /; hence we need only consider homogeneous functions of (a, /3, y, 8), that is to say quaternary quantics of the coordinates. Any linear function of the coordinates represents a plane ; a = is the plane BCD ; whilst / = is the plane at infinity. Any quaternary n-tic represents a surface of the nth degree ; and any ternary n-tic of three coordinates represents a cone. Also since the equation of a sphere is S + u = 0, where S is a given sphere and u an arbitrary plane, the equation of any sphere may be expressed in the form S + Iu = 0, which shows that all spheres intersect the plane at infinity in the same circle. This circle is of course imaginary, and corresponds to the circular points in plane geometry. - 4. We shall usually employ the symbols Un, Un to denote ternary quantics of (/3, r. (vii) The first polar of every point passes through each conic node; and the axis of every binode has tritactic contact with the first polar. The first part follows from (13) by putting Ui = 0, and differen- tiating with respect to D, which may be any arbitrary point; in like manner the second part follows by putting w, = Los = a beak, elSoi' = appeared. 22 THEORY OF SURFACES a triple point* at A, the tangents at which are given by the equation V3 = 0. There are accordingly three primary species of unodes, according as the triple point is of the first, second or third kinds ; and we shall show that their respective constituents are 0=3, B = 0; (7=2, 5 = 1; 0=1, B = 2. Also since the characteristics of a unode on a quartic surface are the same as on any other surface, we shall employ the surface (30). in which A is the unode and ABG the uniplane. Writing down the discriminant of (30) regarded as the binary cubic (B, 1)^ = 0, it follows that the tangent cone from D is of the 10th degree, and that the term containing the highest power of a is aJVa, which shows that DA is a triple generator of the first kind on the cone. Accordingly the number of tangent planes which can be drawn through DA to the surface is m— Q. The tangent cone from A is (S^Fo + SB'V, + SBV, + Vsy = l2B'{S'W, ■+ SB'W, + SBW, + W,) (31), which is of the 6th degree and has three nodal generators which are the lines of intersection of the planes B — 0, F3 = ; hence this cone is of the 24th class; and therefore 24 tangent planes can be drawn to it through DA. But each of these 24 tangent planes is a tangent plane to the surface ; accordingly m — 6 = 24, giving TO = 30. Also by § 15, the class of an anautotomic quartic surface is 36, and therefore the reduction of class produced by a unode is 36 — 30 = 6. When two of the tangents at the triple point on the section by the uniplane coincide, we may take Vs = ^^0. Now X is a small positive quantity when A' lies between A and B, and negative when A' lies on the side of A remote from B ; if therefore the preceding inequality is satisfied when X is posi- tive, the planes will be real ; but they will be imaginary when X is negative. 38. The tangent plane at a pinch point touches the first polar. Let .4 be a pinch point and ABC the tangent plane, then the equation of the surface is j^n-2g2 ^ ^n-z (^^^ + V3) + . . . = 0, and the first polar of D, which is any arbitrary point, is 2a'^-2S + a'^-s {^v\ + v\) + . . . = 0, where v'n —■ dvn/d8, which proves the theorem. 39. A pinch point, in common with all singular points of the same character, is an incident of a nodal line or curve, which NODAL LINES 29 cannot be created nor annihilated by assuming any relations be- tween the constants of the surface, although such relations may apparently make some of them disappear by causing them to coincide. It is of course possible to make the nodal tangents at an arbitrary point coincide, but this does not introduce an additional pinch point but alters the character of the nodal line or of the surface. For example, the equation of a cubic surface on which ^i? is a nodal line and A and B pinch points is i9a7' + 2/332 + ^3 = (34 a). Let the plane a' = a — A-^S = cut AB in B' \ then referring the surface to the tetrahedron of reference AB'CD, (34 a) becomes pay"" + yS ipXy' + qS') + ^/g = 0. If the cubic has a third pinch point at B', p = or q= 0, in either of which cases (34 A) becomes a cuspidal cubic cone. 40. Every tangent plane to a nodal line of the first kind touches the surface at n— 2 points. The equation of the surface may be written in the form QoOL^'-'yS + a**-^ {/3 {P,y^ + Q^yS + R,S') + v,] + ... + /3'-^ {Pn-.y' + Qn-.yB + Rn-^^') + . . . = 0. Writing herein a' + X/S for a, the coefficient of ^^~^ is QoX^'-'yB + v"-' (Ar + QiyB + Ri^') +• • • Pn~2y'+Qn-.yB+Rn-28'=o. Hence, if \ be determined by the equation V-^R, + V-'R^ +... Rn-2 = 0, the plane y will touch the surface at n — S other points, which together with A make n — 2 points. Similarly the plane 8 touches the surface at the point A and the n — S points determined by the equation \^-'P, + X^-'P, + ... Pn-^ = 0. 41. The equation of a surface having a nodal line of the third kind is of the form («, ^T-H, + (P, Q, R, S^y, Sf = 0, where P, Q, R, S are quaternary quantics of all the coordinates of degrees n — S. The line has no pinch points, but it has n—2 cubic nodes or triple points, which are determined by the equation (a, ^y-^ = 0. If A be one of these points, the first term must be 30 THEORY OF SURFACES of the form (a, fiy~^0V2, which shows that there is a tangent cone at A whose equation is of the form ^^^2 + ^3 = 0, and is therefore a cubic cone, on which AB is a nodal generator. 42. If three surfaces of degrees I, m, n intersect in a straight line, which is a multiple line of orders p, q, r on each of them respectively; then the number of their ordinary points of intersec- tion is Imn — Iqr — mrp — npq + 2pqr. In order to prove this theorem, we shall employ a method very successfully used by Salmon*, which depends upon the principle that the number of points of intersection of three surfaces is an invariable quantity ; in other words a point of intersection cannot be created nor annihilated by means of any relations between the constants, or by making the surfaces degrade into improper ones. We may therefore replace the surface Si by p planes Up passing through the given straight line, and another surface Si^p which does not contain the line. Treating each of the other surfaces in the same way, we have to find the number of points of intersection of the compound surfaces Si-.pUp, S^-qUq, and Sn-rUr which do not lie in the line. Now Si-p, Sm-q and Sn-r intersect in {I — p)(m — q) {n — r) points; also Si^p, Sm-q and Ur intersect in (I —p){m — q)r points; hence the total number of points of intersection which do not lie in the line is (f,—p){m—q){n—r)-{-{l—p){m — q)r + (m — q)(n — r) p + (n — r){l — p)q = Imn — Iqr — mrp — npq + 2pqr (35), which shows that the number of points absorbed by the straight line is Iqr + mrp + npq — 2pqr. If the three surfaces have a common straight line, p = q = r = I, and the number of their ordinary points of intersection is Imn — l—m — n + 2 (36). If however the line is a nodal line on the surface I, and an * Camb. and Dublin Math, Journ. vol. ii. p. 65. NODAL LINES 31 ordinary line on the other two, p = 2, q = r=l, and the number of points is Imn - I - 2m - 2n + 4 (87). 43. A nodal line of the first or second species on a surface of the nth degree reduces the class by 7n — 12. Since a nodal line on a surface gives rise to an ordinary line on the first polar, we must put I = m = n — l; p = q = l, r=2; and (35) becomes n(n-iy-5n + 8. This would give the reduction of class were it not for the pinch points ; but we have shown in § 38 that the tangent plane at a pinch point touches the first polar, hence every pinch point absorbs an additional point of intersection. Accordingly the total number of ordinary points of intersection is n(n-iy-5n + 8-(2n-4^) = n (n - If -7n + 12, and therefore the reduction of class is 7n—12. It follows from the preceding investigation that before the results of § 42 can be employed to determine the reduction of class produced by a singular line on a surface, it is necessary to examine the intersection of the surface and its first polar at the singular points on the line. Hence the above result is not true in the case of a nodal line of the third kind, since such lines possess cubic nodes but no pinch points. 44. If a surface of the nth degree possesses a multiple line of order n — 2, it has 2 (3n — 4) other lines lying in it. If AB be the multiple line, the equation of the surface is OPVn-i + O-^Wn-^ + ^(Tn-2 + °t-Vn-i + ^^n-i + Vn = 0, and the section of the surface by the plane B = ky consists of AB repeated n — 2 times and the conic (A,B,G,A',B',G'^a,^,yy = 0, where A, B, C are polynomials of k of degree n — 2; A', B' of degree n — \; and G of degree n. If the plane is a tangent plane, the conic must degrade into a pair of straight lines, the condition for which is that its discriminant should vanish, which furnishes an equation of degree 3n — 4 in k. Hence the surface contains twice this number of lines. 32 THEORY OF SURFACES 45. If an anautotomic surface contains a straight line, the number of tangent planes which can he drawn through it is (n + 2){n- 2)1 If AB be the line, the equation of the surface is a^'-^Vi + a""-^ {I3wi + Wj) + . . . = 0, hence ^5 is a line lying in the first polar of every point on the line. Accordingly by (36) the number of points of intersection of the surface and its first polars with respect to A and B which are absorbed by the line is Sn — 4, and therefore the number of remaining points is n{n - ly - Sn + 4! = (n + 2)(n - 2f. Either of these theorems show that 10 lines intersect every line lying in a cubic surface. On the Intersections of Surfaces. 46. Three surfaces of degrees I, m, and n intersect in Imn points, but when three or more surfaces are given by a set of determinants, it frequently happens that they possess a common curve; and we shall now show how to determine the degree of this curve. If u, u ; V, v' ; w, w he quaternary quantics of degrees I, m, n respectively, the degree of the common curve of intersection of the surfaces included in the set of determinants u , V , w = u', v', w' is mn ■{■ nl -\- Ini. The three surfaces are vw' = wv, wu=uw', uv' = vu' (38), and are of degrees m + n, n + l and l + m respectively ; and the first and second surfaces intersect in a curve of degree (m + n){n + I). But this curve is a compound curve consisting of the curve of intersection of the surfaces w = 0, w' = 0, which is of degree n^, and a residual curve of degree (m + n) (n + I) — n"^ = mn + nl + Im. The coordinates of points on the curve w = w' = obviously do not ON THE INTERSECTIONS OF SURFACES 33 satisfy the third of (38) ; hence the residual curve alone is the one common to the three surfaces. 47. The four surfaces included in the set of determinants u, u' , u", u'" =0, w, w', w", w" where the us, v's and w's are of degrees I, m and n respectively, possess a common curve of intersection of degree P + m? + n^ + mn + nl + Im. Let A = vw — wv, B = wu' — uw, G = uv' — vu' . . .(39), then the surfaces formed by omitting the fourth and third columns respectively are Au" +Bv" +Cw" =0' Au'" + Bv"' + Gw"' = 0' ..(40). The two surfaces (40) are each of degree l + m + n, and their curve of intersection consists of the common curve of intersection of the three surfaces A =0, B = 0, C = 0, which has been shown to be of degree mn + nl + Im, and a residual curve of degree l^ + m^ + n'^ + mn + nl + Im. Now if we write down the identity Au+Bv' + Gw' = (41), and eliminate A, B and G from (40) and (41), w^e shall obtain the determinant formed by omitting the first column, and the deter- minantal surface thereby formed will be satisfied by all values of the coordinates which satisfy (40) but which do not make A, B and G vanish ; that is to say the coordinates of all points on the residual curve. Hence the determinantal surface formed by omitting the first column contains the residual curve. In like manner by writing down the identity Au + Bv + Gw = (42), and eliminating A, B and G between (40) and (42), it can be shown that the determinantal surface formed by omitting the second column also contains the residual curve. Hence the four surfaces included in the set of determinants intersect in a common curve of the degree above mentioned. B. 3 34 THEORY OF SURFACES 48. The six surfaces included in the set of determinants = 0, U, V, w, t U, V, w , t where the u's, v's, w's and t's are quantics of degrees I, m, n and p, intersect in Imn + mnp + npl + plm common points. The determinant is formed by eliminating the constant k between the four equations u = ku', v = kv'] .(43), w = kw', t = kt', and the condition that (43) should be satisfied by the same values of the coordinates is that the elirainant of (43) should vanish. Now it is known* that the highest power of k in the eliminant is equal to Imn + mnp 4- npl + plm ; hence this is the number of sets of values of the coordinates which satisfy (43), and therefore the determinantal surfaces intersect in this number of common points. When u, v, &c. are planes, the surfaces consist of six quadrics which possess four common points of intersection. This may be verified as follows. The three quadrics formed by omitting the last column intersect in a twisted cubic curve ; and the quadrics uv = vu' and ut' = tu intersect in the line u = 0, v = and a second twisted cubic, and the points of intersection of the two twisted cubics are those common to the system. Now if three quadric surfaces possess a common straight line, it appears from (36) that the latter absorbs four out of their eight points of intersection ; hence the two twisted cubics intersect in four points, Avhich are the ones in question. The Hessian. 49. The Hessian of a surface is the locus of points whose polar quadrics with respect to the surface are cones. The locus of the vertices of these cones is a second surface called the Steinerian. * See Salmon's Higher Algebra, 4th edition, § 78. If P, Q, R, S be four quaternary quantics of degrees I, m, n, p respectively, the eliminant is a homo- geneous function of degree mnp of the coefficients of P ; of degree npl of those of Q ; of degree plm of those of R ; and of degree linn of those of S. THE HESSIAN 35 Let us temporarily employ (^, rj, f, w) to denote current coordinates ; then the polar quadric of any point (a, /3, 7, S) is Let a=^, f=dm' ^ = ci«^'^'-'^'- •••^^^^' then (44) may be written in the form ap + 6772 + c^2 + c?«2 + 2/^^ + 25r^f + 2/1^77 + 21^00 + 2m77ft) + 2;i^a> = 0. . .(46), or in the abbreviated one (a, h, G, d, f, g, h, I, m, n\^,7],^,aif = (47). The discriminant of (46) may be expressed in the form of the symmetrical determinant* H = a, h, g, I (48), h, h , f, ra g> f , c, n \ I, m, n, d the vanishing of which expresses the condition that (46) should reduce to a cone. If the original surface is of degree n, each of the constituents of H are of degrees n — 2; hence the degree of the Hessian is 4 (w - 2). The Hessian may be expressed in a variety of forms, one of the most convenient of which is the following. Let A=bc-f\ B = ca-g\ G = ab - h') •(49), A' = gh-af, B' = hf-hg, G'=fg-ch] also let A be the determinant formed by erasing the last row and column in H ; then H=M-{A, B, G, A\ B', G'Jl, m, ny (50). The determinant obtained by putting d — Q is a well known one, since it expresses the condition that the straight line la + wyS + 717 = should touch the conic (a, 6, c,f,g,h^a,^, 7)' = 0; * The letter n is used in two different senses in this and the following sections ; but the reader will find no difficulty in avoiding confusing them. 3—2 36 THEORY OF SURFACES hence the last term of (50) equated to zero is the tangential equation of the conic. 50. The Hessian passes thfough every double point ; also its curve of intersection with the surface is the spinodal curve. The equation of a surface passing through A is a^-^Ui + a^-^U2+...Un = (51), and the polar quadric of A is {n-l)(xUi + U2 = (52). When J. is a double point Mj = and the polar quadric reduces to the nodal cone, which shows that the Hessian passes through A. When -4 is a point on the spinodal curve, we may put u^ = S, in which case u, = 8\ + 8(p^ + qj) + rrf, and (52) becomes [{n - 1) a + 8wo +^/3 + ^7} 8 + ^7^ = 0, which is the equation of a quadric cone. This proves the second part and shows that the degree of the spinodal curve is 4/i {n — 2). 51. Every conic node on a surface gives rise to a conic node on the Hessian, having the same nodal cone and the same lines of closest contact. We shall choose the tetrahedron of reference so that the equation of the surface is C^n-2 (^^2 ^ ^^g) ^ an-3j^^ +...Un = (53), the advantage of which is that when p = the singularity at A becomes a binode, and when g* = it becomes a unode. Retaining only the leading terms, we obtain a = {n-2)(n- 3) (p^ + qryB) a''~*\ b = 2pa'^-^ + a^-^d^Us/d^"" c = a'^-H'Us/dy'' d = a''-'d'us/d8^ f= a^'-^d^'Us/d/Sdy g = (n—2) qa^~^S h = 2{n- 2) j9a"-3yg l = {n-2) goL^-^y m = a'^-H'Us/d^d8 n = qa""-^ + a^'-^^'us/dy dS y .(54). THE HESSIAN 37 Substituting these values in (50) it will be found that the only terms involving a^n-io^ which is the highest power of a, are which is equal to 2pq' (n - 1) (n - 2) (pyS^ + qyS) a^"-" (55), which proves the first part of the theorem. To prove the second part of the theorem, we must calculate the coefiicient of a^**"", which is rather a long expression and seems hardly worth while writing down. The result is as follows. Let AG and AD he two of the lines of closest contact, then Us = /3« + /S^ (\y + fjiB)-\-/3 (Ly^ + 2MyB + M') + yS (Fy + G8), and if U3 be the coefficient of a^''^-" in the Hessian, it will be found that AG and AD are generators of the cone U^. 52. The spinodal curve has a sextuple point at a conic node. If m and n are the degrees of two surfaces, which intersect at a point A which is a multiple point of order ^ on one surface and of order q on the other and the nodal cones are not specially related to one another, any plane section through A will consist of two plane curves which have multiple points of orders p and q at A. These curves will therefore intersect in mn—pq ordinary points, which shows that J. is a multiple point of order pq on the curve of intersection of the two surfaces. If however p = q and the nodal cones are identical, it can be shown as follows that the order of the multiple point on the curve is ^ (^+ 1); for consider the surfaces a^'-PUp + a''-P-'Up+, + ...Un = 0, where n > m. Multiply the second equation by a^-^ and subtract from the first and we obtain a^'-P-' (up+, - Up+,) + . . . w„ = 0, which is the equation of a surface passing through the curve of intersection of the two surfaces and having a multiple point of order p + 1 &,t A, whence ^ is a multiple point of order ^ (^ + 1) on the curve. In the case of a surface and its Hessian p =2, and p{p + 1)=6. 53. Every binode on a surface gives rise to a cubic node of the third kind on the Hessian, two of the tangent planes at which 38 THEORY OF SURFACES coincide with the biplanes ; also the spinodal curve has an octuple point at the binode. When ^ = 0, the double point becomes a binode, and (55) vanishes. In this case the highest power of a is the (4w— ll)th, and the terms containing it are 2bgnl - abn' = {n-l){n-2) a'^'-^^yBd^Us/dlS^ and consequently J. is a cubic node on the Hessian whose nodal cone consists of the two biplanes and the plane d^u^jd^^ = 0. Such a singularity is called a cubic node of the third kind. To prove the last part consider the two surfaces a"-3;S7S + a"-%4 +...«/« = 0, where n> m, from which we deduce which shows that A is an octuple point on the curve of inter- section of the first two surfaces. 54. Every unode on a surface gives rise to a quartic node on the Hessian, whose nodal cone consists of the uniplane twice repeated and a quadric cone. When q = 0, the singular point becomes a unode, and the term containing the highest power of a is which is a quartic node of the species described. 55. If a straight line lie in a surface, it will touch the Hessian and therefore the spinodal curve. Let AB be the line, let A be one of the points where AB cuts the Hessian, and let ABG be the tangent plane at A ; then since the section of the surface by the plane 8 must have a cusp at A, it follows that the equation of the surface must be 7 {a«-2 (p7 + qh) + a'^-^w^ + . . .} + h{oi^-^U, + (f'-^U^ + ...)==0 (56). Now when 7 = S = 0, the values of a, b, h are zero ; and the Hessian reduces to {fl — gmf = ; and on calculating this expres- sion it will be found that yS^ is a factor, which shows that the line AB touches the Hessian at A. THE HESSIAN S9 56. The curve of contact of every trope on a surface forms part of the spinodal curve. Let a be the trope, (a, Og) the curve of contact, B any point on it, then the equation of the surface is a^Uo + a'^-'Un-i + . . . a^ (/Q'^-Vo + yS'^-Vj + . . .) + ai^-'To + ^^'-'r, + ...) + n/(/3"-^«Wo + ...) = 0, where fl^ = ^'-\ + ^'-% + .... The polar quadric of 5 is a (ao-o + (w - 1) /3to + Ti] + Vi^Wo = 0, which represents a cone. The reader is doubtless aware that Fresnel's wave surface possesses 16 tropes whose curves of contact are circles, and that 4 of these tropes are real whilst the remaining 12 are imaginary ; hence the spinodal curve consists of these 16 circles, which make up an improper curve of the 32nd degree. 57. If a fixed plane touches a surface along a straight line, it also touches the Hessian along the same line. Hence the line twice repeated forms part of the spinodal curve. The equation of the surface must be of the form a"-^a + a'*-^ (B\ + Bv^ + ry^) where ^5 is the line B the fixed tangent plane and J. is any point on the line ; also the suffixed letters are binary quantics of (yS, 7). Forming the Hessian, the coefficient of a^**"^ will be found to be AB, where -4 is a constant. 58. If a surface has a nodal curve of degree b and a cuspidal one of degree c, the degree of the spinodal curve is 4?^(7^-2)-86-llc. This theorem is due to Cayley*, who proved it in a different manner ; and it will be sufficient to consider the case of a surface having a nodal line, the equation of which is a'^-^yB+a'^-^Us+...=0, where u^ = ^fu + 78^ + B^w, u, V and w being arbitrary planes through A. Now if we proceed in the same way as in | 51, it will be found that the highest * C. M. P. vol. VI. p. 342. 40 THEORY OF SURFACES power of a in the Hessian is a*"~^^ and the term involving it is (If+mgy. Omitting a, the coefficient is = {n- 2)2 q^ (P72 + QyS + RBJ, whence -45 is a quadruple line on the Hessian having two pairs of coincident tangent planes ; accordingly AB repeated 8 times forms part of the spinodal curve. In the same way it can be shown that if the surface has a twisted nodal curve of degree b, this curve is a quadruple curve on the Hessian, and that the former 8 times repeated forms part of the spinodal curve. Hence the degree of the residual intersection of the surface and its Hessian, which is the true spinodal curve is 4w (n — 2) — 86. The equation of a quartic surface having a cuspidal line will hereafter be shown to be (pay + q^By + (P, Q, R, S^y, Bf = 0, where P, Q, R, S are arbitrary planes, and by forming the Hessian by the same method it will be found that the line AB repeated 11 times is part of the spinodal curve, but the work is rather long. 59. In the same paper Cayley has also proved the corre- sponding theorem for the flecnodal curve which is : — If a surface has a nodal curve of degree b and a cuspidal one of degree c, the degree of the flecnodal curve is w(llw-24)-226-27c. A similar theorem undoubtedly exists for the bitangential curve, but so far as I am aware it has not been investigated. 60. The Hessian of an anautotomic surface of degree n possesses 10(n—2y nodes*. Let us for convenience suppose the original surface to be of degree n + 2, so that each of the constituents of the Hessian is of degree n ; let dH ^ dH ^ dH ^ dH ^ ,^., M=^' u=-^- a^=-^- -Tn=-^^ ('^)' * Cayley, Proc. Lond. Math. Soc. vol. in. p. 23. C. M. P. vol. vii. p. 133. THE HESSIAN 41 also let J)= h , f, m (58), f , c, n in, n, d then by means of (49) and (50) the following additional equations can easily be proved, viz. : (59), .(60). BV'-AA' = Af, BC - A'' = Aa, &c. A =Aa + B'g + G'h \ A, = Al +G'm+B'n \ A,= C'l + Bm + A'n A3 = B'l + A'm + Cn Eliminating I between the second and third of (60) and taking account of the first, we obtain A^A — AjC" = A (wc - nf). Hence if Q is any point at which two of the three deter- minants A, Aj, A2 vanish, the third determinant will also vanish, provided none of the quantities A, C and mc — nf vanish at Q. Now the equations A = C' = mc — nf= are equivalent to hlg = b/f=f/c = m/n (61), so that the exceptional points, which we shall denote by P, are the common points of intersection of the six surfaces included in the set of determinants h, b, f, m =0 (62), g> / c, n and since each constituent is of degree n, it follows from § 48 that there are 4n^ points such as P. The system of 4?i^ points included in (62) lies on the Hessian and also on the surface i) = ; for if we put each of the ratios (61) equal to k, and substitute in the determinants for ^and D, they obviously vanish. We have therefore proved that if Q be a point such that dHjdd = and dH/dl = 0, then dH/dm = and dH/dn = 0, provided Q does not coincide with any of the 4/i^ points P ; also the points at which these differential coefficients vanish lie on the Hessian. 42 THEORY OF SURFACES In the next place write H in the form d, I, m, n I , a, h, 9 m, h, b, f I , g, f , < and let dH/dc = 0; then since dH/dn has been shown to vanish it follows that dH/dg = dH/df = 0. Proceeding in the same way it can be shown that : If any two of the differential coefficients of the Hessian with respect to the four letters a, b, c, d vanish; and any one of those with respect to the six letters f g, h, I, m, n also vanishes, then all the differential coeffcients will vanish, provided the points at which they vanish are not included amongst the points P. In explanation of the proviso, let the three equations be dH da = ^=0 ?^=0 ' dd ' di (63), then these equations will represent three surfaces of degrees Sn which intersect in a certain number of points, or which may possess a common curve ; and if Q be any point common to the three surfaces all the differential coefficients of H will vanish at Q unless Q coincides with any of the 4/?.^ points P. We shall now show that the points Q are nodes on the Hessian. Let a = ttoO.^ + aitt""^ + . . . , &c. ; where an={^, y, BY; then TT TT A„ fdS dH J + ... Let the tetrahedron be chosen so that A coincides with one of the points Q, then Hq and all its differential coefficients vanish, and the highest power of a is a.*^^~^, which shows that A is a node on the Hessian. From these results it follows that the nodes are included amongst the intersections of the surfaces D = 0, and a, h, 9> I h, b, f, m 9> /. c, n = .(64). THE HESSIAN 43 Now it follows from § 47 that the four determinantal surfaces (64) intersect in a common curve of degree 6n^ ; also the 4)^' points P lie on this curve and also on the surface D ; and we have to show that the points P are not nodes, and that the number of the points Q is equal to lOri^ Let A be one of the points P, which are the points common to the surfaces (61), then it follows that &c., &c. Hence by (49) A = (b, + c,-2f,)a'^-\ 5 = (ao-l)a^ (7 = (ao-l)a^ ^' = - (ao - 1) a^ B' = (h, +f, -b,- g,) a^-\ G' = (A + g^-(h-h,)a^-\ Substituting in H and D we obtain H= {{ao-l)do - lo'+2k] {b, + c,-2f,) a'^-' + ... , D = (cZo-l)(6, + c,-2/0a^"-^+.... The first equation shows that J. is a point but not a node on the Hessian, and the second equation shows that the Hessian and D touch one another at -4. Now the common curve of intersection of (64) obviously lies in the Hessian ; and since it is of degree 6/i^ it intersects the surface D in 18w^ points and touches it at the 4?i^ points P; hence the number of remaining points of intersection, which are the points Q, is equal to 18w^ — 8w.^ = lOn^, which is therefore the number of nodes on the Hessian. 61. When the polar quadric degrades into a cone, the locus of the vertex of the cone is a surface called the Steinerian. This surface is the analogue of the plane curve of the same name, which is the locus of the points of intersection of the pair of straight lines constituting the polar quadric of points on the Hessian. The Hessian* and the Steinerianf are discussed in the papers referred to below. * Hesse, Grelle, vols, xxviii. and xli. ; Sylvester, Phil. Trans, cxliii. ; Del Pezzo, Rendiconti, Napoli, 1883 ; Brill, Math. Annalen, vol. xiii. ; Segre, Rendiconti, Lincei, 1895. t Crelle, vol. xlvii. CHAPTER II CUBIC SURFACES 62. There are altogether twenty-three different species of cubic surfaces, which depend upon the number and character of the point singularities which they possess. The class of an anau- totomic cubic surface is n (71 — ly = 12 ; the degree of the tangent cone is n(n — l) = 6; also the cone has n(n — l)(n — 2) = 6 cus- pidal generators, but no nodal ones. From this it follows that a cubic surface cannot have more than four nodes ; for if it had five nodes, the class of the surface would be two, and quadrics are the only surfaces of the second class. 63. Every cubic surface can be expressed in either of the forms Su=S'u' (I), or a^j = uvw (II), where S, S' are quadric surfaces, and a, ^, 7, u, u , v, w are planes. We have shown that every cubic surface possesses straight lines lying in it ; hence if AB be one of these lines, the equation of the surface must be of the form S = 0, u =0; also the plane u' = 0, which contains the conic, intersects the cubic in the straight line u = 0, u' = 0. The equation 8' = \8, where \ is an arbitrary constant, represents any other quadric surface which intersects the cubic surface in the quartic curve ; and if 8 and 8' be eliminated we obtain the equation u = \u', which shows that the residual curve of intersection is a conic lying in the plane u = \u', which passes through the straight line w = 0, u' = 0. The corresponding theorem for a plane cubic curve led to Sylvester's discovery of the Theory of Kesiduation of curves ; and it will hereafter be shown that the preceding theorem is a particu- lar case of a corresponding Theory of Residuation of surfaces. 65. Through any straight line on a cubic surface, 5 planes can be drawn which intersect the surface in a pair of straight lines. Also 27 straight lines can be drawn on an anautotomic cubic surface. The first part of this theorem has already been proved in §§ 44 or 45 ; and the second part can be established as follows. Let any triple tangent plane intersect the cubic in the lines X, fji, v; then since four other planes can be drawn through X, each of which intersects the cubic in two other straight lines, these four planes will furnish eight straight lines which together with \ make nine. Similarly, each of the other lines fi and v will furnish nine more, making a total of 27. 66. An anautotomic cubic surface has 45 triple tangent planes. Each of the 5 tangent planes drawn through a straight line on a cubic touches the latter in three points, which are the vertices of the triangle which is the section of the surface by the plane ; and since 27 lines can be drawn on the surface, the number of such planes is 27 x 5 = 135. But every plane such as ABG contains each of the lines AB, AG and BG; hence the total number of distinct planes is 135 -^ 3 = 45. _ 46 CUBIC SURFACES 67. An anautotomic cubic surface has 54 tangent planes at which the point of contact is a tacnode on the section. The section of the surface by an arbitrary plane through the line AB consists of AB and a conic which in general cuts AB in two distinct points, and the plane is therefore a double tangent plane ; but if the two points coincide, the point of contact is a tacnode on the section. Let the equation of the surface be + ja^Mo + a (pl3 + v^) + jS^Wo + I3w, + ^2)8 = (1). In (1) write S = ky, divide out by 7, and then put 7 = 0, and (1) becomes a'(Uo- Jcuo) + a/3 (P-pJc) + fiHWo- kw,) = (2). Equation (2) determines the two points in which the line AB cuts the conic ; but if AB touches the conic the two roots of (2) must be equal, which furnishes a quadratic equation for deter- mining k. Hence on each of the 27 lines there are two points such that the section by the tangent plane consists of the straight line and a conic touching it at the point; accordingly there are 54 of such points. 68. The literature on the 27 lines of a cubic surface, and the different ways of arranging them is voluminous ; and the reader is referred to the authorities cited below*. One arrangement is that of a double-six, which consists of two systems of lines 1, 2, 3, 4, 5, 6, 1', 2', 3', 4', 5', 6', such that each line of one system intersects every line of the other system except the line represented by the figure above or below it, as the case may be. There are altogether 36 double- sixes. The Hessian of a Cubic Surface. 69. We shall now give some theorems relating to a cubic sur- face and its Hessian, which is a quartic surface, since 4 (w — 2) = 4 when w = 3. * Schlafli, Phil. Trans, cliii. (1863), p. 193 ; Cayley, Ihid. clix. (1869), p. 231, and G. M. P. vol. vi. p. 359, vol. vii. p. 316 ; Clebsch, Math. Annalen, vol. iv. ; Cremona, Crelle, vol. lxviii. ; Dixon, Quart. Jouni. vol. xl. p. 246 ; Burnside, Ibid. p. 381. THE HESSIAN OF A CUBIC SURFACE 47 If the polar quadric of any point A with respect to a cubic he a cone whose vertex is B, the polar quadric of B is a cone whose vertex is A. Hence the Hessian and the Steinerian are identical. This theorem has already been given in § 13 (viii) ; but for a cubic surface the proof is as follows. Let the equation of the surface be a^ + ahi^ + au2 + U3 = (3), then the polar quadric of A is SaP + 2au, + U2 = (4), and if (4) is a cone whose vertex is B, it cannot contain /3 ; hence Ui = ]j,y + vS, U2 = V2, and (3) becomes a^ + a^ (fiy + vB) + av2+ Us = (5). The polar quadric of B is now du^jd^ = 0, which is a cone whose vertex is A. The points A and B are points on the Hessian, and are called conjugate poles ; and a theory exists with reference to them analogous to the corresponding theory for plane cubic curves. 70. The tangent plane to the Hessian at A is the polar plane of B with respect to the cubic. The polar plane of B with respect to the cubic is d^u^/d^'^ = ; and if we write down the Hessian of (5) it will be found that the only terms which contain a^ are b (acd + 2gln - an^ - cP - dg'^) (6). Now b = d^U3ld^^, whilst the term in brackets is equal to where -4 is a constant ; hence the equation of the tangent plane to the Hessian at A is d^u^jd^'^ = 0, which proves the theorem. Let V2=^72 + 2g7S + rS2 (7), u^ = ^X + /S' (% + i\^S) + /3 (P72 + 2Q7S + R^) + Frf + SGy'8 + SHyB' + K8' (8), then the tangent planes to the Hessian at A and B are Sl3wo + My + m = •(9). 3a + fly +vS =0 71. If the line joining two conjugate poles A and B intersect the. Hessian in two points A' and B', the tangent planes to the 48 CUBIC SURFACES Hessian at A and B will intersect in the line joining the two poles which are conjugate to A' and B'. Let /3' = /3 — Xa = be the plane A' CD ; then since the points G and D are arbitrary, we may suppose that G is the pole con- jugate to A'. Changing the tetrahedron to A'BGD, (5) becomes a^ + a^ (fiy + vB) + av^ + (\a + ^J w, + (Xa + ^J ( % \ M) + (Xa+ yS0(-P7'+ 2Q7S + l'')-vFrf + ... = (10). The polar quadric of ^' is a cone whose vertex is G, and must not therefore contain 7 ; hence /i = 0, ilf=0, ^ + XP = 0, g + XQ = (11). Let D be the pole conjugate to B' , and change the tetrahedron to AB'GD, where a' = a — X'/3 = is the equation of the plane B'GD, then the condition that the polar quadric of B' should be a cone whose vertex is D will be found to lead to a system of equations, which when combined with (11), give i;=0, N=0, q=Q = 0, E = -XV (12). Equations (9), which are the tangent planes to the Hessian at A and B, now become /3 = 0, a = 0, which intersect in the line GD. 72. The polar plane with respect to the cubic of any point on the line joining two conjugate poles, passes through the line of intersection of the tangent planes to the Hessian at these points. By the last section the equation of the cubic is a'+a(py+rB') + ^'Wo + ^{Pj' + R8') + w,= (13), and the polar plane of any point (a', /3') on AB is which passes through GD. 73. The polar plane of the Hessian, with respect to any point on a cubic, intersects the tangent plane to the cubic at that point, in the line which passes through the three points of inflexion of the section of the cubic by the tangent plane. Consider the cubic a?^ + a {^v, + ySvi + 78) + ^'w, + ^'w^ + j3w,^-rf ^h'=0.. .(14). This surface passes through the point A, and /3 = is the tangent plane thereat; also the section of (14) by the plane /3 is 078 + 73+8^ = (15). THE HESSIAN OF A CUBIC SURFACE 49 Equation (15) is a cubic curve which has a node at A, and the line CD passes through the three points of inflexion on the section. Let then if we write down the Hessian of (14) it will be found that the only terms which involve a^ and a^ are - abn^ + h^n^ - 2ghmn - 2hfnl = 4,a^ (a + 2Q/3 - 2pq/3). The polar plane of A with respect to the Hessian H is d'Hjda' = 48 (2a + Q^ - pqjS) = 0, which passes through CD. 74. If the polar quadric of a point A consists of a pair of planes, then A is a conic node on the Hessian. The polar quadric of A is in general a cone whose vertex is B ; and since the line AB must possess at least one degree of freedom, the equation of the polar quadric of A must contain at least one variable parameter \. If therefore X is made to satisfy the con- dition that the discriminant of the polar quadric of A should vanish, the cone will degrade into a pair of planes whose line of intersection passes through B. From (5) and (7) the polar quadric of the cubic with respect to J. is 8a2 + 2a {^l'y + vh) ^-p^^ + 2^70 4- rS^ = (16), and its discriminant equated to zero gives 3pr + 2qiMv — Sg-^ — pv"^ — r/^^ = (17). Equation (6) gives the term involving a^ in the Hessian, and if the values of a, c, &c., be substituted from (5) it will be found that (17) is the condition that this term should vanish. Hence A is a conic node on the Hessian. 75. Let BG be the line of intersection of the planes, then (16) must reduce to ^a? + 2a.vB + r82 = 0, which requires that fju- p = q = 0, and (5) becomes a'+va'S + roiS'+Us = (18), and since B may be any point on BG, it follows that the polar quadric of every point on BG consists of a cone whose vertex is A ; hence BG is a line lying in the Hessian. B. 4 50 CUBIC SURFACES Let us now enquire whether the Hessian has any nodes lying on BG. Let C be any point on BG, and let /3' = /3 — X7 = be the equation of the plane A G'D ; then in order to find the polar quadric of G', we must change the tetrahedron to ABG'D and differentiate (18) with respect to 7. Accordingly the polar quadric of 0' is where u^ is given by (8). Writing this out at full length we obtain + 2(Q\+SG)yB + 2{N\+Q)B^ + 2(MX + P)^y = 0...{19), and its discriminant when equated to zero furnishes a cubic equation for X, showing that there are three points on BG at which the polar quadric degrades into a pair of planes. Hence the Hessian has three conic nodes lying on BG. 76. Every anautotomic cubic surface can be expressed in the canonical form aoL^ + 6/3^ + C72 + dh^ + eu^ = 0, where a+/3 + 7+S + M=0. The preceding form of a quaternary cubic is due to Sylvester*, ■ and we shall now show how it can be established by means of the foregoing results. Let J. be a node on the Hessian ; then we have already shown that (i) the polar quadric of A consists of a pair of planes inter- secting in a straight line LMN] (ii) the line LMN lies in the Hessian ; (iii) it has three conic nodes upon it, which we shall suppose situated at the points L, M, N. Let a = 0, u = be the equations of LMN; then the polar quadric of A must be of the form aa^ — eu^ = 0. Integrating with respect to a, it follows that the equation of the cubic surface must be of the form C7 = aa^ + b^ + 07^ + dh^ + eu^ + ^8^ (/i7 + vh) + yS (P72 + 2Q7S + M^) + Gy'^h + Hyh^ = . . .(20). Let us construct a triangle bv drawing three lines LBD, MGD and NGB in the plane a through L, M and N. Then each of * A quaternary cubic can also be expressed as the sum of six cubes. See Dixon, Proc. Lond. Math. Soc. Series 2, vol. vii. p. 389. THE HESSIAN OF A CUBIC SURFACE 51 these lines has one degree of freedom in the plane a ; also since a may be any plane which passes through LMN, the plane a has also one degree of freedom. The coordinates of L may therefore be taken to be a = 0, 7=0, ^ + 8 = 0; hence the polar quadric of X is dl3 d8 ~ ' which by virtue of (20) is 36/32 + 2/3 (fiy + vS) + P72 + 2Qy8 + E8' - SdB' - 7/32 - 2yS (Q7 + BB) - Gry' - 2ir7S = . . .(21). This is the equation of a cone whose vertex is A ; but since L is by hypothesis a node on the Hessian, it follows that (21) must degrade into a pair of planes intersecting in some line passing through A, which we may take to be AG, since AG may be any line passing through A. This requires that f^ = p = Q^G = H = 0, which reduces (20) to U= aa' + bfi' + cy' + dB' + eu' + v^h + R^S' = 0. . .(22). The coordinates of M are a = 0, /3 = 0, 7 + S = 0; and its polar quadric is Scy^-3dB'-vl3^-2R^B=0 (23), which is a cone whose vertex is A. Now A and N are fixed points, and the line ^C is a fixed line, hence AGN is a fixed plane, but AB may be any line through A in this plane. We shall therefore suppose it to coincide with one of the generators of the cone (23), in which case v = 0. Also M is by hypothesis a conic node on the Hessian, hence the discriminant of (23) must vanish, which requires that Be = 0. Taking B = 0, (22) reduces to U = aa? + b^' + cy' + dB^ + eu' (24), the form of which shows that JSf is also a node on the Hessian. 77. To find the equation of the Hessian. Let (a', /3', 7', h') be any point on the Hessian, u' the corre- sponding value of u ; then the polar quadric of this point with respect to the cubic is fdV dU\ ^,iW dU\ 4—2 52 CUBIC SURFACES which by virtue of (24) becomes aa'a^ + 6/S'^2 ^ ^y^s ^ ^g'g2 ^ g^'^2 = q (25). Since (25) is a cone, its discriminant must vanish ; and the latter is obtained by eliminating (a, y8, a/3yS (51). This surface is called Steiner's* quartic, and its properties will be discussed in the chapter on Quartic Surfaces. A cubic having four conic nodes is also the envelope of the quadric (A, B, G, F, G, H~^\, fj,, vf — 0, where A, B ... are planes. * Crelle, vols, lxiii. and lxiv. CHAPTER III TWISTED CURVES AND DEVELOPABLES 94. Every curve, plane or twisted, may be regarded as the limit of a polygon whose angles are denoted by the figures 1, 2, 3, &c. The lines 12, 23, &c. are tangents at successive points on the curve, and the plane 123 contains these lines. When the curve is twisted, the point 4 will in general lie in a different plane 234 which intersects the plane 123 in the line 23; also since these planes pass through three consecutive points on the curve they are osculating planes. A developable surface is a ruled surface each generator of which intersects the consecutive one ; hence the envelope of the osculating planes to a twisted curve is a develop- able surface. Since a point which is constrained to move along a twisted curve has only one degree of freedom, the osculating plane has likewise the same degree of freedom ; hence the constants in its equation must be functions of a single parameter 6. Accordingly the equation of the osculating plane must be of the form z = ex + y(f>(d) + y}r(e) (1), where ^ and -x/r are arbitrary functions*. 95. When a developable surface is generated in the above manner, the curve whose osculating planes envelope the develop- able is called the edge of regression ; and when the curve is an algebraic one, the edge of regression is a cuspidal f curve on the developable, that is to say the surface consists of two sheets which touch one another along a cusp. If a sheet of paper is * The elimination of 6 leads to the two well-known partial differential equations q = (j) (p) and rt = s^. t The Italians call a cusp un reijresso ; a rhamphoid cusp un regresso di seconda specie and so on. TWISTED CURVES AND DEVELOPABLES 65 bent along the lines 12, 23, &c., the continuations of these lines in the opposite directions will form a twisted curve on the developable ; but if pieces of fine wire be gummed to the paper along the lines 12, 23, &c., whilst the continuations are left free, the latter portions will generate the other sheet of the developable. The surface therefore consists of two sheets, one of which is the bent paper, whilst the other is generated in the manner above described. The point where any plane section of the developable cuts the edge of regression is a cusp on the section ; and it is obvious that if it were possible for a developable to consist of one sheet only, the point in question would be a point d'arret, and such a singularity cannot be possessed by an algebraic curve. Any plane through a tangent, such as 12, cuts the developable in a curve having a point of inflexion at 1 and also in a generator which is a stationary tangent to the curve. If a point moves along the tangent from left to right keeping in the same sheet of the developable, the point will begin to move along the curve as soon as it has passed 1 ; for the continuation of the tangent is a generator on the other sheet of the developable. 96. Every generator of a developable of degree v is cut by v — 4i other generators. Any plane through a generator 12 intersects the developable in this generator and in a curve of degree v — 1 ; and since the generator 12 touches this curve at a point of inflexion, it cuts it in V — 4> points which are the points where y — 4 other generators cut 12. 97. The reciprocal polar of a twisted curve is a developable surface amd vice versa. Let and P, lying in the plane of the paper, be the origin of reciprocation and any point on the curve; and let the osculating plane at P be perpendicular to the plane of the paper and cut it in PA. Draw OA perpendicular to the osculating plane at P and produce it to A', so that OA . OA' = ]S', aS" are quadric surfaces and u, v are planes. Eliminating 7, B we obtain Sv = S'u (3). The two cubic surfaces (3) and the first of (2) intersect in a twisted quartic of the first species, which is the complete inter- section of the quadrics S and 8\ and in a residual quintic curve which lies in both the surfaces (2) ; accordingly although (2) are the simplest equations for defining the curve, it may be expressed as the partial intersection of two surfaces of higher degrees. But the Theory of Residuation furnishes the simplest method of dis- covering the two surfaces in question. 103. We must now consider the characteristics of E and D. The degree w of ^ is equal to the number of points in which the curve intersects an arbitrary plane. Now the class of a curve is defined to be that particular geometrical property which is the reciprocal of its degree ; and, in the case of a twisted curve, this property is the number of tangent planes which can be drawn to D' through a point. But every tangent plane to D' is an osculating plane to E' ', hence the class m of a twisted curve is equal to the number of osculating planes which can be drawn to it through a point. A node B can arise in three ways : (i) when the two surfaces containing the curve have ordinary contact at a point on the curve ; (ii) when one surface passes through a double point on the other; (iii) when one of the surfaces possesses a nodal curve. The reciprocal of a node is a doubly osculating plane ot; and such a plane is a double tangent plane to the developable D'. Since ib- must intersect the curve in at least six points, it cannot SINGULARITIES OF TWISTED CUKVES 69 occur on any curve of lower degree than a sextic ; but, unlike a double tangent to a plane curve, it need not occur at all. Let the osculating plane at a point P cut the curve in points Q, R, S, ... ; then since P has one degree of freedom, its position is determined by means of a single parameter 6; accordingly the distance QR can be expressed as some function F of 6. If, therefore, the para- meter of P is a root of the equation F{0) = O, the points Q and R will coincide and the plane will touch the curve at Q ; hence there is a determinate number of osculating planes which touch the curve elsewhere*; but a third point 8 cannot, in general, be made to coincide with Q without assuming some relation between the constants of the curve. The plane tn- is equivalent to the two osculating planes at its points of contact. A cusp K occurs when the surfaces containing the curve have stationary contact at some point on it. A cusp may also, like a node, arise from the surfaces having singularities. The reciprocal of a cusp is a stationary plane a. Such a plane passes through four consecutive points on the curve, and at the point of contact two osculating planes coincide. The point is also one at which the tortuosity of the curve vanishes and changes sign ; and since the tortuosity is some function / of 0, the plane a occurs at the points whose parameters are the roots of the equation f{0) — 0. This plane osculates the developable D' along a generator. Since a cusp is the reciprocal of a stationary plane, the pre- ceding argument shows that a cusp is a point through which four consecutive osculating planes pass. If, therefore, the developable is the envelope of the plane aa -f- 6/3 + C7 + (ZS = 0, where a, h, c, d are functions of a single parameter 6, the conditions for a cusp are obtained by differentiating the above equation three times with respect to 6 and eliminating (a, ^, 7, 8). A double tangent t to E reciprocates into a double tangent t' to F'. This singularity, unlike a double tangent to a plane curve, need not occur ; for if a straight line touches a twisted curve, it need not intersect it elsewhere. The double tangent is a generator of D ; but since the osculating planes to E at its points of contact are in general distinct, there are two tangent * These planes will be discussed in § 111. 70 TWISTED CURVES AND DEVELOPABLES planes to D along a double tangent to E. Hence a double tangent to ^ is a nodal generator on D. A stationary tangent t to E reciprocates into a stationary tangent l to E'. Such a tangent touches the curve at a point of inflexion, which is a point where the curvature vanishes and changes sign ; hence l may always occur on a twisted curve and must never he assumed to he zero. A stationary tangent to ^ is a cuspidal generator on D. It will hereafter appear that a considerable number of twisted curves do not possess stationary planes and tangents, the expla- nation of which is as follows. The tortuosity and curvature can always be expressed in the form F {d) ; but if none of the roots of the equation F {6) = are the parameters of points on the curve, the tortuosity or curvature (as the case may be) can never vanish. A simple example is furnished by the expression for the curvature of an ellipse in terms of the excentric angle 0, which cannot be made to vanish for any real value of ^. There are four other singularities which have to be taken account of. If be the vertex of any cone which stands on the curve, a generator which intersects the curve in two points P and Q is a nodal generator of the cone and gives rise to an apparent node on the curve* ; since, to an eye situated at 0, two branches of the curve appear to intersect one another. The number h of such nodal generators is equal to the number of apparent nodes on the curve. The reciprocal singularity is called an apparent douhle plane, and consists of a pair of osculating planes whose line of intersection lies in a fixed plane. The number g of such pairs of planes is equal to the number of apparent double planes. It is also possible for a pair of tangents to E to lie in a plane, which is consequently a double tangent plane to E ; and the locus of the points of intersection of such pairs of tangents is a nodal curve on D. The degree x of this curve is equal to the number of * Since a " double point " includes a cusp as well as a node, the phrase "apparent double point" is inappropriate; for a twisted curve cannot, in general, possess an apparent cusp. By properly choosing the vertex of the cone, it is quite possible for the curve to appear to have a cusp to an observer situated at the vertex ; but if the vertex be shifted, the apparent cusp will be changed into an apparent node. THE PLUCKER-CAYLEY EQUATIONS 71 points in which it intersects an arbitrary plane ; accordingly x is equal to the number of pairs of tangents to E, whose points of -intersection lie in a fixed plane. The reciprocal singularity y is equal to the number of pairs of tangents to E', which lie in planes passing through a fixed point ; in other words, it is equal to the number of double tangent planes to the curve which pass through a fixed point. The Plucher-Gayley Equations. 104. Thirteen quantities have therefore to be considered ; and the reader will observe that I have assigned definite geo- metrical meanings to each of them, so that there is no need to employ such verbose and obscure phrases as "rank of the system," " planes through two lines," and the like. For the purpose of facilitating comparison, I subjoin my own notation and that of Cay ley and Salmon*. Basset — v, n, in, 8, ct, k, a, r, t, h, g, x, y. Cayley and Salmon — r, m, n, H, 0, /3, a, co, v, h, g, x, y. Cayley obtained equations connecting the thirteen charac- teristics of the curve in the following mannerf. Consider any plane section 8 of the developable D. Let jB, Jtt, ^, ^, 'S^, 31 denote the degree, class, and number of nodes, cusps, double and stationary tangents to S. The degree iS, of >Sf is equal to the number of points in which it is cut by an arbitrary line lying in its plane ; that is to say to the degree of D. Hence ^ = v. The class JW of >S is equal to the number of tangents which can be drawn to it through a fixed point in its plane ; and since each tangent lies in a tangent plane through to D, this number is equal to the class of D, that is to the class of E. Hence JW = m. A node on S may arise in two ways, (i) If two tangents TP, TQ to E intersect the plane of /Sf in a point T, the osculating planes to ^ at P and Q intersect the plane of S in two lines * The symbols 8, k, t, i have been used in England for many years to designate the number of nodes, cusps, double and stationary tangents which a plane curve possesses ; and to employ different symbols for the corresponding singularities of a twisted curve introduces confusion and unnecessary complexity. t C M. P. vol. I. p. 207 ; Liouville's Journ. vol. x. p. 245. 72 TWISTED CUHVES AND DEVELOPABLES TP', TQ', which are tangents to S at T. Hence two branches of S cross one another at T, and T is a node. The number of nodes arising from this cause is equal to the number x of pairs of tangents to JS, whose points of intersection lie in the plane of S. (ii) In the next place the point where every nodal generator of D cuts the plane of ^ is a node on S; and since every such generator is a double tangent to E, there are t nodes on S arising from this cause. Hence "^ = cc + r. A cusp on S also arises in two ways, (i) The points where the plane of S cuts E are obviously cusps on S, and their number is equal to the degree n of E. (ii) In the second place every stationary tangent to E gives rise to a cuspidal generator on D, and the points where these generators cut the plane of S are cusps on S; there are consequently i cusps arising from this cause. Hence Wi = n + t. A double tangent to S arises in two ways, (i) If P and Q are two points on E, the osculating planes at which intersect in a line pq lying in the plane of S, and Pp, Qq are the corresponding generators of D, the line pq is a double tangent to S whose points of contact are p and q. There are accordingly g double tangents arising from this cause, (ii) In the next place every doubly osculating plane 'ur to E intersects the plane of 8 in a line which is a double tangent to 8. Hence '^T = ^ + -ot. A stationary tangent to 8 can only arise from the existence of stationary tangent planes to E ; hence E = cr. We must therefore write in Pliicker's equations and we obtain TO = y (j^ - 1) - 2 (a? + t) - 3 (n + t) ' o- = 3z/ (i/ - 2) - 6 (a; + t) - 8 {n + i) V = m (m — 1) — 2 (^ + ot) — 3cr n + 1 = 3??i (m -2) — Q{g + zT)~8(T of which only three are independent. Four more equations can be obtained by considering the cone which stands on the curve; but although the use of the cone is instructive as a mathematical artifice, its employment is un- necessary, since the four remaining equations can be obtained from (4) by writing for each quantity its reciprocal. We thus obtain .(4), THE PLUCKER-CAYLEY EQtTATlO;b^S 73 ■(5), )i = v(v-\)-2(y+T)-S(m + i) \ ^ = Sv(p-2)-6{i/ + T)-8(m + i) v = 7i(n-l)-2 {h +S)-3k m + t = Sn (n - 2) - (5 {h + 8) - 8k of which only three are independent. 105. It may be wortli while to give a direct proof of (5). The degree of the cone is equal to that of the curve, and its class is equal to the number of tangent planes which can be drawn through any arbitrary line passing through its vertex ; and since the vertex may be any arbitrary point, this is equal to the number of tangents to E which intersect an arbitrary line, that is to the degree of D. Hence ^ = n, ifWl = v. Every generator of the cone which passes through an actual or an apparent node is a nodal generator. Hence 3© = A + S. A cuspidal generator can only occur when the curve has a cusp. Hence 1£t = «• A double tangent plane to the cone arises in two ways, (i) When the curve possesses a pair of tangents which lie in a plane passing through the vertex, and the number of such pairs of tangents is y. (ii) Every plane through the vertex and a double tangent to ^ is a double tangent plane to the cone. Hence A stationary tangent plane to the cone also arises in two ways. (i) Every tangent plane to D, which passes through the vertex of the cone, is an osculating plane to E, and the number of such tangent planes is m. (ii) In the second place every tangent plane which passes through a stationary tangent to E is likewise a stationary tangent plane. Hence 5 = tti + t. Substituting in Plticker's equations for the cone we obtain (5). 106. There are certain other equations, called the Salmon- Cremona equations, the consideration of which will be postponed for the present, and we shall proceed to find the characteristics of the curve of intersection E of a pair of surfaces U and V whose degrees are M and iV respectively, and which are arbitrarily situated with respect to one another. The degree v of the developable D is equal to v = MN{M + N-2). 74 TWISTED CURVES AND DEVELOPABLES .(6). ■(7), ,(8). The degree of D is equal to the number of tangents to E which intersect an arbitrary straight line. Let the equations of the latter be poL+ q/3 + ry + s8 = OJ The equations of the tangent to E are where CTj = dV/d^, &c., (^, rj, f, w) being the point of contact ; and the condition that (6) and (7) should intersect is P, Q, R, S, p, q, r, s, u„ u„ u„ u„ Fx, F„ V,, V,, Equation (8) is a sui-face of degree M + N—2, which passes through the points of contact of those tangents to the curve which intersect the line (6), and since the number of such points is equal to the number of points of intersection of U, V and (8), the former is equal to MN(M + N-2). If (^, 7), ^, co) be the coordinates of any point in space, (7) are the equations of the line of intersection of the polar planes of with respect to the two surfaces ; and (8) shows that if this line intersects a fixed straight line, the locus of is a surface of degree M + N-2. 107. The degree of the curve is obviously equal to MN; also if the surfaces do not touch one another B= k = 0, whence, by the third of (5), 2h = MN{M-l)(N-l) (9). Now the number of apparent nodes is obviously independent of the number of isolated singularities of the curve ; hence the above value of h is true when the curve is autotomic. If, therefore, we substitute the value of h from (9) in (4) and (5), these equations may be reduced to the following six : — v = MN(M + N-2)~2B-Sk (10), m = nMN{M+N-S)-68-8K-i (11), a = 2MN(SM+SN-10)--128-15K-2i (12), INTERSECTION OF TWO SURFACES 75 2{g + m) = MN [QMN{M^N - 3)^ - 22 (if + i\^) + 71} - QMN (M + N- 3) (6S + 8« + i) + {6B+ 8k + tf + UB+ 5Qk + 1i (13), 2{x + t) = MN {MN (M + N -2y - 4^ (M + N) + 8} -2MN(M + N-2){2S + Sk) + (2S + 3/c)- + 8S+ll«-2i (14), 2 (2/ + t) = MN {MN(M+ N - 2f - 10 (if + N) + 28} - 2 JfiV^ {M + N-2) (2B + 2k) + (28 + 3«)^+208+27«: (15). We have already shown that, when the surfaces containing the curve are anautotomic and are arbitrarily situated with respect to one another, S = k = 0; but Salmon has assumed, without proof, that t = 0, which is not permissible. That t is zero may be proved as follows. In order that the curve may have a point of inflexion at P, the tangent at P must have tritactic contact with both surfaces. Let PT be the tangent at P, and let PM, PM' be the nodal tangents at P to the section of one of the surfaces by its tangent plane at P; and let PN, PN' be the corresponding tangents at the node on the section of the other surface by its tangent plane at P. Since P has only one degree of freedom, it is always possible to determine its parameter so that one or other of the tangents PJf, Pilf' shall coincide with PT ; but since a second equation of condition is necessary, in order that one or other of the tangents PN, PN' should coincide with PT, a stationary tangent cannot exist unless the surfaces are special ones, or are specially situated with reference to one another. It is also evident, from the discussion in § 103, that ot and t cannot occur unless some special conditions are introduced. 108. If the curve of intey'section of two surfaces is an irredu- cible one, the surfaces cannot touch one another in more than \MN{M+N-^) + l points. Let S be the maximum number of points of contact ; then 3 + A is the maximum number of nodal generators which any cone standing on the curve can possess ; whence 2 (S + A) = {MN - 1) {MN- 2). Substituting the value of h from (9), we obtain the required result. 76 TWISTED CtTRVES AND DE7EL0PABLES 109. We shall now show that when the complete curve of intersection of two surfaces degrades into a pair of irreducible curves of degrees ih and n^, the characteristics of one curve can be found when those of the other are known. We shall suppose that neither of the surfaces have stationary contact with one another, in which case neither curve can have any cusps; but if the two curves intersect, their points of intersection will be nodes on the compound curve and the two surfaces will touch at these points. Let the suffixes 1 and 2 refer to the two curves ; and let H and h' be the number of their apparent and actual intersections. Then 7i,+ n,-=MN (16), and, by (9), 2{h, + h, + H) = MN{M-l){N-l) (17). Applying the third of (5) to each of the curves 1 and 2, we obtain Vi= ^1 Oh — 1) — 2/ii1 ; , .; \ (18). 2/3 = ^2(^^2- l)-2/io| whence, taking account of (16), we obtain v^-v,^(n,-n,)(MN-l)-2(h,-h,) (19). Applying the same equation to the compound curve, we obtain V, + v, = MN{MN- 1) - 2(A, + /?, + fi"+ h'). Substituting the values of v^, vo. from (18), and taking account of (16), we obtain* H+h' = n^n. (20). Equation (20) determines the number of actual intersections of the two curves, when the number of apparent intersections is known, and vice versa. The polar planes of U and V, with respect to a point 0, inter- sect in a line i, and we have shown in § 106 that if L intersects a fixed straight line, then will lie on a surface of degree M -\- N —2. At a point where this surface intersects the curve ??i, L will be a generator of the developable enveloped by its osculating planes ; but if be a point of intersection of the curves n^ and n.2, the two planes will coalesce with the common tangent plane to the two surfaces, and L will be the line joining to the point where the tangent plane cuts the fixed line. Hence * Otherwise thus. The common generators of the wo cones standing on the two curves must pass through the apparent and the actual intersections of the two curves. PARTIAL INTERSECTION OF TWO SURFACES 11 Substituting the values of v-^, v.^ from (18) and that of 8' from (20), we obtain 2/.,+ ff = n,(i/-l)(iV^-l)| 2/i2 + ^ = w,(if-l)(iV^-l)J ^ ^' accordingly 1{]i^-h^ = (n^-n^{M-\){N -\) (23), also from (21) v^-v^ = {n^-n^){M+N-1) (24). Since the surfaces are supposed to be so situated that i is zero, we obtain, in like manner, from the last of (5) m^-m^ = Z{n^-n^{M-\-N-^) (25), and from the first two of (4) o-i-o-2=2(wi-??2)(3if+3iY-10) (26). The preceding equations contain the principal formulae, and the reader can easily extend them to the case in which the two curves have actual nodes and cusps. 110. When the vertex of the cone standing on the curve has a special position, its characteristics are different from those of a cone whose vertex is arbitrary ; and the following table, due to Cayley*, gives various results of importance. ^j , T^ /M Nodal Cuspidal m i Stationary Vertex Degree Class Generators ditto ^^^^^^^ ditto 1. On a tangent n v-\ A-l + S k + 1 j/-i' + 4 m — 2 2. On the curve n—\ v -2 h-n + 2 k y — 2u + 8 m-3 + 8 +T +1 3. At a node n-2 i/-4 h-2n + 6 k i/-2p + 20 m-6 8-1 +r +t 4. At a cusp n-2 v-3 h-2n + 6 k-1 ^-3v + l3 to-4 + S +T +1 5. On a stationary n v-2 A -2 + 8 k-|-2 y-2v-\-^ m-Z tangent +r +t- 1 6. At the point of n-l v-Z h-n + l « + ! y-3i^ + 14 m-4 contact of ditto +8 +r +t— 1 7. On a double n v-2 A- 2 + 8 k+2 ?/-2j' + 10 m-4 tangent + t — 1 + 1 S. At a point of n-l v-S h-n + 1 k + 1 y-3i/ + 15 m-5 contact of ditto +8 + t — I + 1 * G. M. P. vol. VIII. p. 72 ; Quart. Joiirn. vol. xi. p. 294. 78 TWISTED CURVES AND DEVELOP ABLES The reciprocal of the cone is a plane section of the developable D', and the corresponding characteristics of the section are its class, degree, double tangents, stationary tangents, nodes, and cusps. Also in the eight special cases, the plane : — 1 passes through a tangent ; 2 is a tangent plane ; 3 is a double tangent plane ot ; 4 is a stationary plane cr ; 5 passes through a stationary tangent l ; 6 is the tangent plane at contact of ditto ; 7 passes through a double tangent t ; 8 is a tangent plane at one of the contacts of ditto. In each of these eight respective cases, the singularities of the section can be obtained from the table by writing for each quantity its reciprocal. Denoting, as before, the characteristics of the section by old English letters, it follows from Pllicker's equations that if the values of ^, Jtt and 31 can be found, the remaining three can be found from the equations 21B = M'-^OM + 8iW - SI, Wi = sM-sM + ^, 2^ = in(iB-l)-ia-33I. To prove 1, we observe that since the plane passes through a tangent to E, the section of D consists of the tangent and a residual curve of degree v—1. Hence ^=1^ — 1. The class of the section is equal to m ; hence Jtt = m. The tangent through which the plane passes is a stationary tangent to the section, hence B = o- + 1. Substituting in the first equation, we obtain 2'm=v^-l2v + 8+8m- Sa, and from the first two of (4), we obtain 8m - 3o- + j;2 = lOi; + 2 (x + t), which gives iIB = ic + T — y + 4, and the rest may be proved in a similar manner. The Salmon-Cremona Equations. 111. We have already shown that a determinate number of planes exists, which osculate the curve at one point and touch it at another. The reciprocal singularity, which we shall call y, consists of a point the tangent at which intersects the curve. Let the tangent at P intersect the curve at Q ; then Q is a point THE SALMON-CREMONA EQUATIONS 79 where two tangents intersect on the curve, and therefore the nodal curve x intersects E B,t Q. Every double tangent plane to E contains a pair of tangents which intersect on the nodal curve x. The envelope of these planes is a developable called the hitangential developable ; and its class is equal to the number of double tangent planes which can be drawn through a fixed point, that is to y. Its reciprocal polar is the nodal curve on the developable D'. Since every double tangent plane has one degree of freedom, every curve possesses a determinate number t' of triple tangent planes ; and the tangent lines at the three points of contact form a triangle whose vertices lie on the nodal curve on D. The reciprocal polar of a plane t' is a point on the nodal curve on D' at which there are three tangent planes ; in other words it is a cubic node on D'. The point may also be regarded as one from which three tangent lines can be drawn to E' ; or as a triple point on the nodal curve on D'. Let k be the number of apparent nodes on the nodal curve on D; then the reciprocal singularity is the number of apparent double planes of the hitangential developable of E'. Let q be the class of the nodal curve on D ; then its reciprocal polar is the degree of the bitangential developable of E'. We have therefore the following additional eight quantities, the last four of which have a reciprocal relation to the first four. 7, the number of tangents which intersect E in one other point. t, the number of triple points on the nodal curve x. k, the number of apparent nodes on x. q, the class of the nodal curve x. y, the number of tangent planes which osculate E at one point and touch it at another. t', the number of triple tangent planes to E. k', the number of apparent double planes of the bitangential developable of E. q', the degree of the latter developable. Cay ley*, who was in correspondence with Cremona at the time, has given eight equations connecting these quantities with * G. M. P. vol. VIII. p. 72. 80 TWISTED CURVES AND DEVELOP ABLES the first thirteen characteristics of the curve; but instead of following Cayley's method I shall employ that of Zeuthen*, who made use of united points in the Theory of Correspondence. This will illustrate a totally different method of dealing with these and other problems. 112. The Theory of United Points, which is due to Chaslest, depends upon the following theorem. On a given straight line let there he two sets of points, whose distances froin a fixed point on the line are x and y. Let X points X correspond to a given point y ; and let fi points y corre- spond to a given point x. Then if the distances between the two sets of points are connected by an algebraic equation, the number of points X luhich coincide with points y is \+ /j,. These sets of coincident points are called united points. By hypothesis, the distances between the two sets of points are expressed by means of an equation of the form x^ (Ay + Byi^-^ +...) + ^^~^ {^'y'^ + B'yi"-^ +...) + ... = .. .(1), for if y has a determinate value b, equation (1) is of degree A, in ^ ; whilst a X has a determinate value a, (1) is an equation of degree [X in y. When x = y, (1) becomes an equation of degree \-\- jxinx, which proves the theorem. Cayley+ has extended this theorem to curves of deficiency p, and has shown that : — If two points on a curve of deficiency p have a (X, fi) correspondence, the number of united points is X + /jb-{- 2kp where k is a constant to be determined. But it will not be necessary to consider this extension of Chasles' theorem. Since the reciprocal of a point on a fixed line is a plane through another fixed line, Chasles' theorem is true in the case of two sets of planes through a fixed straight line, which have a (X, fi) correspondence. It frequently happens that p points x coincide with q points y, in which case the point is equivalent to pq united points ; for if we select any single point x of the group, there are q points y which coincide with it ; and since there are p coincident points x, * Annali di Matenuitica, vol. iii. Serie 2, p. 175. t Nouvelles Annales de Mathematiques, 2" Serie, vol. v. p. 295. X C. M. P. vol. VI. p. 9 ; Proc. Lond. Math. Soc. vol. i. April 16, 1866. SCROLL ENVELOPED BY A TRISECTANT 81 the total number must be pq. The numbers p and q are deter- mined by the conditions of the problem under consideration. A line which cuts a twisted curve three times, four times, &c., is called a trisecant, quadrisecant, &c. We are now in a position to proceed with the proof of the Salmon-Cremona equations. 113. The trisecants envelope a scroll whose degree T is T={n-^)[h-:Ln{n-\)]... (2). Let X be the distance of any point on a fixed line L from a fixed point on L ; let the line xPQ cut the curve in P and Q ; let RPy and RQy' be two lines through another point R on the curve, which cut L m y and y' ; and take x and y, y' as corre- sponding points. Since the plane xFy cuts the curve in n points, there are n — 2 points such as R ; also since each of the lines RF and J?Q gives rise to a ?/ point, the plane xRy produces 2 (?i — 2) points of type y corresponding to a single point of type x. But the number of lines such as «PQ, which can be drawn through a single point x, is equal to h ; hence the total number of points y which correspond to a single point x is 1h {n — 2). Accordingly \ = ^ = 2/i(?z-2) (3). United points will occur : — (i) When the point R coincides with Q. In this case Qy' is a tangent to the curve, and since there are w — 2 points of type x in the plane through L and Q, this plane produces n — 2 united points ; but z/ tangent planes can be drawn through any fixed line to the curve, hence the total number of united points is v{n— 2). (ii) When the plane through L contains a double point. The degree v of D, which is equal to the number of tangent planes that can be drawn through L to the curve, is given by the equation v = n{n- l)-2h-28-SK (4); accordingly when the curve is anautotomic, the value of v is given by the first two terms ; and (4) shows that every plane which passes through a node is equivalent to two tangent planes, and every plane which passes through a cusp is equivalent to three. From this it follows that when the curve possesses B nodes and K cusps, the number of united points is equal to (n — 2) (28 + 3/c). B. 6 82 TWISTED CURVES AND DEVELOP ABLES (iii) When the line xPQ is a trisecant. Let R be the remaining point in which this line cuts the curve ; then since the trisecant is a triple generator of the cone whose vertex is x, the number of its distinct apparent nodal generators is ^ — 3, and the number of distinct points y produced by them is 2 (w — 2) {h — 3). Also the number of points y arising from the plane through L and the trisecant is 3(?i— 3). But if we take any line yPS in this plane, we shall find that the number of distinct points x corre- sponding to y is 2 {h - 1) (w - 2) + 2 (w - 4) + 2 -f 1 = 2/i {n - 2) - 1, which shows that the point in which the trisecant cuts L is equi- valent to two points X ; and therefore each of the points y under consideration is equivalent to two points, making a total of 6 (w — 3). Accordingly the total number of united points pro- duced by the trisecant is 2 (71 - 2) /i - 2 (w - 2) {h - 3) - 6 {n - 3) = 6, and if Ttrisecants intersect L, the total number is QT. Collecting our results we obtain \ + fi = (v + 28+SK)(n-'2.) + 6T. Substituting the value of X+ /j, from (3), and that of 28 + 3/c from (4), we finally obtain T=(n-2){h-in(n-l)] (5). 114. The degree of the bitangential developable is q' = v{n-S)-2B-SK (6). Let P and Q be the points of contact of a double tangent plane to E ; then PQ is a generator of the bitangential develop- able and the degree q' of the latter is equal to the number of lines such as PQ which intersect a fixed line L. We must first find the number of points I on L, from which a pair of straight lines lying in a plane u through L can be drawn, each of which intersects the curve E in two points. The plane u will cut E in n points P, Q, ... P^ ; the line PQ will cut L in a point x; the lines joining the n—2 remaining points will cut L in i(?i — 2){n — 3) points y ; and we shall take x and y as corresponding points. Since the cone, standing on the curve, whose vertex is x has h apparent nodal generators, there DEGREE OF BITANGENTIAL DEVELOPABLE 83 are h lines such as xPQ ; hence the total number of points corre- sponding to a single point a; is ^ {n — 2) (w — 3) h. We thus obtain X = ^=^{n-'l){n-Z)h (7). The points I are the only united points ; but since each point I is equivalent to two x points, the former is equivalent to two united points. We thus obtain \+^i = {n-T){n-^)h=n (8). In the second place, h — 1 apparent nodal generators can be drawn through the point x exclusive of xPQ. Let the planes through L and these h — 1 generators be the planes v, and take u and V as corresponding planes. Since the plane w cuts the curve in n points, there will be ^n{n — l) points on L such as x, and consequently there will he \n{n—l){h—l) planes v corresponding to a single plane u ; accordingly \' = ^i' = \n(n-l){h-l) (9). United planes will occur : (i) When the plane u contains a trisecant x'PQR. Since the trisecant is a triple generator of the cone whose vertex is x, the latter has only ^ — 3 distinct apparent nodal generators, and con- sequently there are only ^ — 3 distinct planes v ; hence the trisecant gives rise to two planes v which coincide with u. But the number of distinct points a? on X is now equal to i(7i-3)(n-4)-h3(7i-3) = i?i(7i-l)-3, which shows that three points x coincide with x' ; accordingly a trisecant gives rise to 6 united planes, and since T of them cut L, the total number is %T. (ii) When the plane u contains a generator of the bitangential developable. Let x'PQ be the generator ; then since the tangents to the curve at P and Q lie in the tangent plane to the cone along x'PQ, this generator is an apparent tacnodal generator which is formed by the union of two apparent nodal generators. Hence there are only A — 2 remaining apparent nodal generators, which give rise to one united plane ; accordingly the total number is q. (iii) The v ordinary tangent planes which can be drawn through L to the curve do not give rise to any united planes ; but it is otherwise when a plane u passes through a double point. Let i^ be a node, P any other point on the curve in the plane u, and let NP cut L 'm x; then the line xPN is a triple generator of 6—2 84 TWISTED CURVES AND DEVELOP ABLES the cone whose vertex is x, because the node N gives rise to two tangent planes and the point P to a third, all of which touch the cone along the generator xPN. Let h' be the number of apparent nodal generators exclusive of xPN, h the number of actual nodes on the curve, then // + 3 + 8 — 1 = ^ + S, giving h' = h—2; hence one V plane coincides with a u plane. But since the plane in question is equivalent to two u planes, the node and the point P give rise to two united planes ; also since there are n — 2 points P and 8 nodes, the latter produce 2{n—2)S united planes. (iv) Since a plane through L and a cusp is equivalent to three u planes, it can be shown in the same manner that the k cusps produce 3 (w — 2) «: united planes. (v) The I points mentioned above obviously give rise to 21 united planes. We thus obtain \' + fi' = 6T + q' + {n- 2)(2S + Sk) + 21 Substituting the value of V 4- /a' from (9) and those of T and I from (2) and (8), we finally obtain q' = {n- 3) [n (n-l)- 2h] - {n - 2) (28 + 3«), which by virtue of (4) may be expressed in the form (6) or by the alternative equation q=2h + v{n-2)-n{n-l) (10). 115. The number of tangents which cut the curve in one other point is ry = v{n-4>) + 4^h-2n(n-S)-2i-4^T (11). Let a plane u pass through a line L and cut the curve in n points P. Through one of these points draw a line PQR cutting the curve in Q and R ; also through the line L and the two points Q and R draw two planes v, v' ; and take u and v as corresponding planes. Since by the table to § 110 the cone whose vertex is P has h — n + 2 apparent nodal generators, the number of planes v arising from a point P is 2 {h — n + 2), and since there are n points P, the total number of planes v corresponding to a single plane m is 2n{h — n + 2). Hence X = fi = 2n(h-n + 2) (12). United planes will occur : (i) When a point Q coincides with P, in which ca,se PQR is one of the tangents y. THE SINGULAR TANGENTS T 85 (ii) When one of the points P is a node. The cone whose vertex is P now possesses h — 2n + Q apparent nodal generators, hence the effect of a node is to produce 2(h- n + 2)-2{h-2n + 6) = 2{n-4<) united planes. Accordingly 8 nodes produce 2 {n — 4i) 8 of such planes. (iii) When one of the points P is a cusp. If one of the planes M is a tangent plane to the curve no united planes are produced, but there are 2{h — n + 2) pairs of coincident planes v none of which coincides with a u plane. But a plane which passes through L and a node is equivalent to two tangent planes, and the fact that a node produces 2 (w — 4) united planes indicates that ri — 4 planes v coincide with each of the two u planes. A plane through L and a cusp is equivalent to tliree tangent planes, from which we conclude that a cusp produces 3 (n — 4) united planes making a total of 3 (n — 4) k. (iv) Let one of the points P be the point of contact of a stationary tangent; then by § 110 the number of united planes is 2{h-n + 2)-2{h-n + l) = 2, which makes a total of 2t united planes. (v) Let one of the points P be the point of contact of a double tangent ; then the number of united planes is in like manner 2, but since there are two points of contact this number must be doubled making a total of 4r. (vi) Let the plane u contain a trisecant PQR ; then since this line is an apparent nodal generator on each of the cones whose vertices are P, Q and R, the number of distinct planes v is 2{n-S){h-n + 2) + 6{h-n + l) = 2n(h-n + 2)-6, so that the number of united planes is 6, making a total of 6T. We thus obtain X. + /A = 7 + (w - 4) (2S + 3«) + 2fc + 4t + 6T, which by (12), (4) and (2) may be expressed in the form (11), or by the alternative equation y = n(v + 4<)-Q(v + K)-4<{8 + T)~2t (18). 86 TWISTED CURVES AND DEVELOPABLES 116. The number of triple tangent planes is given by the equation 4,y {v-5) = W + 37' + 7 + 12ct + 48 + 3t (i^ - 6) + 2t + 2t(i/-6)+i^(i^-4)(i;-5) (14). Let Z be a fixed line, a fixed point on it, u an arbitrary point on L. Through u draw a double tangent plane U to the curve, touching it at P and Q ; let TP, TQ be the tangents at P and Q; through TP and TQ respectively draw a seiues of double tangent planes V, V intersecting L in v, v ; and take u and V as corresponding points. The line TP is a generator of the developable D, and it is intersected by y — 4 other generators, hence z/ — 4 tangent planes can be drawn through TP to the curve ; but since one of these is the plane IT, the number of remaining planes which determine the points V is v — 5. Accordingly the number of v points which correspond to a single point u is 2 (y - 5) ; but y double tangent planes can be drawn through the point u to the curve, hence 2y (v — 5) points v correspond to a single point u ; and therefore X = /x = 2y{v-5) (15). United points will occur : (i) When CTis a triple tangent plane t'. Let R be the third point of contact ; then U cuts D in three straight lines and in a residual curve of degree v - S, but since TP touches this curve at a point of inflexion at P, it intersects it in 1/ — 6 remaining points and consequently v — 6 generators of D pass through TP. But TQ and the tangent at R are two of these, therefore the number of remaining generators is v — 8; accordingly the number of distinct double tangent planes, which can be drawn through TP, is v — S. It therefore follows that there are 2 (y — 8) = 2 (1/ — 5) — 6 planes V corresponding to the plane U, hence the number of united points due to the t' planes is Gt'. (ii) When ?7 is a plane 7'. Let this plane osculate the curve at P and touch it at Q ; then the residual curve on D is of degree y — 3 and consequently v — 7 double tangent planes, exclusive of U, can be drawn through TP. Through TQ, v — 4< double tangent planes can be drawn which include U twice repeated by reason of the fact that U osculates the curve at P ; NUMBER OF TRIPLE TANGENT PLANES 87 hence the number of distinct planes Fdue to TQ is v — G, making the total number of points v equal to v — 7+v — 6 = 2{v — 5) — ^. Accordingly each plane y' gives rise to three united points. (iii) When U contains a tangent 7. Let P be the point of contact, Q the point where the tangent intersects the curve ; then the number of planes F is v-6 + v — ^ = 2(v— 5) — 1, so that one united point is produced making a total equal to 7. (iv) When C/" is a doubly osculating plane to E. The degree of the section of D is y-4; also if TP, TQ be the lines of contact of U, each of them osculates the section at P and Q, hence the number of generators which cut TP, exclusive of TQ, is y — 4— 6 — l = z/ — 11. Accordingly the total number of V planes is 2{v — 11) = 2 (i* — 5) - 12 ; and therefore the number of united points is 12ot. (v) When U is an osculating plane at a node P. Let PT, PT' be the nodal tangents ; then the degree of the section is v — 2, and the line PT intersects the section in 3 + 1=4 coincident points at P ; hence the number of generators which cut PT, exclusive of PT', is i/ — 2 — 4-l = i^-7; and as there are two osculating planes at a node, the number of planes V is 2(i/ — 7) = 2(i' — 5) — 4; hence S nodes give rise to 43 united points. (vi) A stationary tangent TP to ^ is a cuspidal generator on D, and the tangent plane to D along it is the cuspidal tangent plane. The degree of the section is therefore 1/ — 3, and the number of generators which intersect TP is y — 3 — 3 = z; - 6. Let one of the planes through TP and a generator TQ be a plane U, then the number of points v due to TP is y — 7, and to TQ is z/-l-3-2 = v-6, making a total of 2z/ - 13 = 2 (z/ - 5) - 3, giving 3 united points. And since the total number of points u is y — 6 and there are l stationary tangents, the number of united points is 34 {y — 6). (vii) We have still to consider the case in which U is the cuspidal tangent plane to D. Since a stationary tangent is equivalent to two ordinary tangents, the point in which U intersects D is equivalent to two u points; also the only generator lying in the plane U is the coincident tangent PT, and there is 88 TWISTED CURVES AND DEVELOPABLES consequently only one v point, so that the total number due to this cause is 2^. (viii) Let a double tangent touch the curve at P and Q ; then since PQ is a nodal generator on D, the degree of any section through D is y — 2, and therefore y — 2 — 3 — 3 = v — 8 other generators intersect PQ ; and if R is the point of contact of one of them with E, the plane PQR is an improper triple tangent plane to E, which is equivalent to one actual triple tangent plane. Zeuthen has omitted the terms due to r, and his method does not apparently enable them to be determined ; I shall therefore assume that each plane such as PQR produces r united points, and that the two osculating planes at P and Q produce s more, making the total number equal to r (z/ — 8) + s, where r and s are positive integers which will hereafter be determined. (ix) Let u be one of the points in which L cuts D. Then if I be the generator through u, v — ^ double tangent planes can be drawn to E through I, and since TJ is one of them the number of Y planes is v — o. But the number of U planes corresponding to the point u is v — 4, which shows that {v — 4) (y — 5) points v coincide with u, and since there are v points u, the total number of united points is v{v— 4) {v -5). Collecting our results we obtain 4?/ (y - 5) = m' + 37' + 7 + 12ot + 48 + 3i (z/ - 6) + 2i ^VT {V -S) + ST + V {v - ^){v - 5). . .(16). In this equation substitute the value of 7 from (13) and the value of 7', which is obtained from (13) by writing for each quantity its reciprocal ; and we shall obtain 6^' = 42/ (z^ - 5) - (3m + ?i) (y + 4) + I80- - 3t (z/ - 8) + 6/c - z; (y^ - 9i; - 4) + {16 - r (i; - 8) - s| T. . .(17). From the Plucker-Cayley equations we easily obtain 3m + n = a -\-^v — i, 2y = v{v- 10) -iSn-SK- 2r, and if these values of Sm + n and 2y be substituted in (17), it follows that all the quantities involved are independent of t except t'. The term involving r is {36 — 4iv — r (v — 8) — s} t, and since we have already shown that the effect of a double tangent is to reduce the number of proper triple tangent planes by v — 8, it follows that 36 - 4y - r (1/ - 8) - s = - 6 (v - 8), NUMBER OP APPARENT DOUBLE PLANES 89 or r (y - 8) + 5 = 2 (y - 6), which gives r = 2, s = 4. Substituting in (16), we obtain (14). Equation (17) may now be reduced to 3^' = -z.(z;-6)(i/-7) + 87i + 2/(3z/-26)-2t (18) by substituting the values of 7, 7' from (13) and reducing by means of the Plucker-Cayley equations. 117. The number of apparent double planes of the bitangential developable is given by the equation 2k' = y(i/-l)-q'-6t'-Sy'-Si{v-6)-2T{v-8)-28-18^ (19). Let Bi) be the bitangential developable, Ef, its edge of regres- sion; and let us denote the characteristics of Ei, by suffixed letters. Then Vi = q', rii^ = y, gi = k', so that the third of (4) of 1 104 becomes 2k' = y (y - 1) - q' - 2^,- Sa, (20). Every double tangent plane to E touches Dj along the chord of contact of the two tangents to E, and is an osculating plane to Eb ; hence if PQ, QR, RP be the tangents to E at the points of contact of a triple tangent plane, the latter will osculate Ei, at three points and will therefore be equivalent to three doubly osculating planes. Accordingly the t' triple tangent planes to E give rise to 2t' doubly osculating planes to E^. Let TPQ be a double tangent to E; P and Q its points of contact ; then v — 8 double tangent planes can be drawn to E through TPQ. Hence if R be the point of contact of one of them, this plane will touch Df, along the generators RP, RQ, and will therefore be a doubly osculating plane to E^. Accordingly t double tangents to E give rise to t{v — 8) doubly osculating planes to Ej). Let PT, PT' be the tangents at a node on E ; then the degree of the section of D by the tangent plane along PT is y — 2 — 4 = y— 6, which shows that PT is intersected by v -Q generators of D, exclusive of PT'. Accordingly the plane TPT' is equivalent to two double tangent planes to E and to two osculating planes to Ej^. Hence this plane is a doubly osculating plane to Ej, and the lines TP, T'P are the tangents at its points of 90 TWISTED CURVES AND DEVELOP ABLES contact ; and therefore B nodes on E produce S doubly osculating planes to Ej,. A doubly osculating plane to E produces on Ej, a compound plane singularity, whose constituents are a certain number of doubly osculating and stationary planes to Ef,. I have not been able to determine the constituents of this singularity directly, and I shall therefore assume that a doubly osculating plane -or to E produces Xzr tangent planes OTi and /liot planes cti to E^,, and we thus obtain 'ST, = St' + T{v-8) + 8 + \'^ (21). Let a plane 7' osculate the curve at P and touch it at Q ; then since every double tangent plane to E is an osculating plane to Ej,, and the plane 7' is equivalent to two tangent planes at P, the former is equivalent to two coincident osculating planes to Eb, that is to a plane v + 19) + 4g — 2m (m — 3) — 2c — 4t, St' = -v{v-6)(v-7) + 8n + y (Su - 26) - 2t, q' = 2h+v{n-2)-n(n-l), 2k' = v^- 9v^ + I7v+y{y-4,iy- = (8), in which the coefficient of ^*"~"^ is zero. This result is important, since it shows that canonical and semi-canonical forms of binary quantics cannot be employed in this subject, since they lead to twisted curves and developables which possess special singu- larities. UNIOURSAL TWISTED CURVES 95 (ii) Let ^ = 0, 6 = QO be the points of contact of a doubly osculating plane ; then the two planes a and Um must be identical, which requires that am= fJ^CL- (iii) Let ^ = 0, 6= co be the points of contact of a double tangent; then since the planes ttTO = and a,^_i = must intersect in the line (a, b) it follows that ajn = pci + qb, cim-i = ra + sb (9). (iv) Let = cc be a point of inflexion, then three consecutive osculating planes must pass through the same straight line, the condition for which is that c = A-a + fib. Transform (1) by writing d = m -1)a + -^a^-i .(15), . .(16), TWISTED CUBIC CURVES 97 which give all the quantities in terms of m, a- and t, and agree with those given by Salmon when cr = t = 0. Twisted Cubic Curves. 128. Every twisted cubic curve is the partial intersection of tiuo quadric surfaces. Every quadric surface contains 9 arbitrary constants, and therefore an infinite number of quadric surfaces can be drawn through 7 points on a cubic curve ; but since a quadric cannot intersect the curve in more than 6 points, it follows that every quadric drawn through 7 points must contain the curve. 129. A cubic curve, which is the intersection of two quadric surfaces having a common generator, cuts all the generators of the same system as the common one in two points, and those of the opposite system in one point. Every generator of a quadric cuts any other quadric in two points which lie on their curve of intersection ; but when the quadrics have a common generator, any generator of the same system does not intersect the common generator and must therefore cut the cubic twice ; but any generator of the opposite system cuts the common generator once, and must therefore cut the cubic once. 130. The most convenient way of representing a twisted curve is by means of the equations of three surfaces which contain the curve; and by §46 a twisted cubic can be represented by the system of determinants = (1), U, V, w u', V, w' where u, u! , &c. are planes. The determinant is equivalent to the system of equations ulu = vjv = w\w' (2), but in practice a simpler method is preferable. Let A and D be any points on the curve, then the cones having these points as vertices which contain the curve are quadric cones ; and by properly choosing the tetrahedron of reference, their equations may be taken to be ^l-rf = 0, a7 - /3^ = 0, B. 7 = : (3). 98 TWISTED CURVES AND DEVELOPABLES from which we deduce aS = ^^, which also contains the cubic. Hence (1) may be replaced by the system of determinants 131. Every twisted cubic is the edge of regression of the developable enveloped by the plane ae' + Wd'+ drye+B = (4). The theory of these developables has already been considered ; and the equation of the envelope of (4) is its discriminant, and is (aS-;g7)^ = 4(a7-^=)(^S-7^) (5), which is a quartic surface. The equations of the edge of regres- sion are obtained from the conditions that (4) should have three equal roots and are a/^ = ^/y = y/8 (6), which are equivalent to (3). The solution of (4) of §121 leads to the three equations a = -^/d=ry/e' = -sie^ hence the coordinates of any point on a twisted cubic curve may be expressed in terras of a parameter by means of the equations a = l, ^ = -e, 7 = ^^ S = -^3 (7), but when we are dealing with more than one point on the curve, each equation must be multiplied by a quantity <^, where ^ is the value of a at each point in question. Accordingly all twisted cubic curves are unicursal. That the cubic curve is a cuspidal curve on (5) may be put in evidence in the following manner. Let A be any point on the cubic and write a + u^, a + v^, a + Wi, a + itj for (a, /3, 7, 8), where the suffixed letters are linear functions of (^, 7, 8) ; then the highest power of a is a^ and its coefficient is (u^ — Sv^ + Swi — tiY, which gives the cuspidal tangent plane at A. 132, If a family ofquadrics have a common curve, the locus of the poles of any fixed plane is a twisted cubic. Let U, V be two given quadrics, and let the fixed plane be oi + /3^y + 8^0 (8), then the equation of any other quadric passing through their curve of intersection is U+\V^O (9). TWISTED CUBIC CURVES 99 Let (/, g, h, k) be the pole of (8) with respect to (9), and let Ui = dU/df, &c. ; then the equation of the polar plane is (10), and if (8) and (10) represent the same plane Eliminating \, we obtain V,-V, V,-V, F3-F4' which are equivalent to (1). 133. All twisted cubics are anautotomic curves. The equations of two quadric surfaces which intersect along the line AB and touch one another at D are (Pa + Q0 + Ry + S8)y + (P'a+Q'^)S = O, (pa + q^ + ry + S8) y + (P'a + Q'^)8 = 0, which shows that they intersect in the line AB and also in the line 7 = 0, P'a + Q'/3 = 0. Hence the residual curve is a conic. 134. The characteristics of the cubic can now be obtained from (14) of § 126 by putting w = 3, k = 0, and are v = 4, m = 3, n = S, h = l, g = 1, and all the other characteristics are zero. Since n = m = 3, it follows that all cubic curves are their own reciprocals in the extended sense of the word, since any point on a cubic corresponds to an osculating plane to another cubic. Also since every twisted cubic possesses one apparent node, it follows that every cone standing on the curve is a nodal cubic cone. Hence every property of a nodal plane cubic curve furnishes a property of a twisted cubic curve ; and this property is capable of furnishing a reciprocal theorem for such curves. 135. Under these cii'cumstances, it seems unnecessary to enter into any detailed discussion of twisted cubic curves; and the following examples will illustrate the method. Let G be the twisted cubic, and >S^ the plane nodal cubic which is the section of any cone whose vertex is 0, which stands upon G. Let P, Q, R be the three points of inflexion of S; then these points lie in a straight line, and consequently in a plane passing through 0. Let the generators OP, OQ, OR cut the twisted cubic 7-2 100 TWISTED CURVES AND DEVELOPABLES in p, q,r; then the tangent planes to the cone along OP, OQ, OR osculate the twisted cubic at p, q, r, and these points lie in the plane OPQR Hence : (i) // from any point three osculating planes be drawn to a twisted cubic curve, their points of contact lie in a plane passing through 0. For a plane nodal cubic curve, the theorem of § 108, Cubic and Quartic Curves, becomes : If AP, AQ be the two tangents drawn from a point A on the curve, and R be the third point where the chord of contact PQ cuts the curve, the tangent at R intersects the tangent at A at a point on the curve. Hence : (ii) Through a point A on a twisted cubic curve and any point draiu two tangent planes OAP, OAQ; and let the plane OPQ cut the cubic in a third point R. Then the tangent planes at R and A which pass through intersect in a straight line which intersects the curve. A plane nodal cubic has 3 sextactic points. Hence : (iii) With any point as a vertex three quadric cones can be described which have sextactic contact with a twisted cubic at three distinct points*. Twisted Quartic Curves. 136. There are two distinct species of quartic curves, the first of which is the complete intersection of a pair of quadric surfaces. These consist of three subsidiary divisions according as the quadrics * Sextactic points on plane curves have been discussed by Cayley, C. M. P. vol. V. pp. 221, 545 and 618, vol. vi. p. 217. He shows that when a plane curve of degree n possesses 5 nodes and k cusps, the number of sextactic points is Hn (4m -9) -245 -27k; from which it can be easily shown by means of Pliicker's equations that this number is also equal to 3m {ivi - 9) - 24t - 27t. On p. 618, some remarks are made with regard to the connection between these points and the reciprocant called the Mongean, see Sylvester's Lectures on Reciproeants. Some further details with respect to plane quartic curves have been given by myself, Quart. Jour. 1903, p. 1. The following is a list of some of the principal memoirs on twisted cubic curves : Mobius, Barycentric Calculus, 1827, Crelle, vol. x. ; Chasles, Apergu Historique, Note xxxiii. ; SchrSter, Crelle, vol. lvi. ; Cremona, Ibid. vols, lviii., Lx.; Sturm, Ibid. vols, lxxix., lxxx., lxxxvi. ; Miiller, Math. Ann. vol. i. The following papers relate to the connection between these curves and the theory of invariants of binary quantics : Beltrami, 1st. Lomb. 1868 ; Voss, Math. Ann. vol. xiii. ; D'Ovidio, Ace. Torino, vol. xxxii. ; Pittarelli, Giorn. di Batt. vol. XVII. TWISTED QUARTIC CUHVES 101 (i) do not touch one another, (ii) have ordinary contact, (iii) have stationary contact ; in which three respective cases the curve is anautotomic, nodal or cuspidal. But t = 0, otherwise a tangent would have tritactic contact with both quadrics, and would there- fore lie in both of them, in which case the curve would degrade into a straight line and a twisted cubic. Similarly r = 0, otherwise a tangent would be a double tangent to both quadrics, which is impossible. Lastly ot = 0, since no curve of a lower degree than a sextic can possess this singularity. Quartics of the second species are the partial intersection of a quadric and a cubic which possess two common straight lines lying in different planes. They cannot possess any actual double points, since as will hereafter be shown, a quadric and a cubic so situated cannot touch one another ; but they may possess one or two points of inflexion, which will occur whenever a generator of the quadric through a point on the curve has tritactic contact with the cubic. This shows that the second species constitutes a totally different kind of curve; also that there are three subsidiary divisions, according as the quartic possesses none, one or two points of inflexion. 137. We shall give for reference a table of the values of the singularities of the two kinds of quartic curves n V m 5 •5J- K <7 T L h 9 X y 7 7' t t' 3 i' h k' 4 8 12 16 2 38 16 8 16 24 60 126 4 6 6 1 4 2 6 6 4 6 4 3 3 4 5 4 1 1 2 2 2 2 6 6 4 6 6 4 3 6' 6 4 4 6 6 6 3 4 6 5 2 1 3 4 5 4 2 6 6 4 3 4 6 4 2 3 3 4 4 6 6 3 3 and the deficiency -p is given by (11) of § 125. It therefore follows that nodal and cuspidal quartics of the first species and all quartics of the second species are unicursal curves, and are therefore the envelopes of the planes (1) of §120, where in the f\yQ respective cases m is equal to 6, 4, 6, 5 or 4. When the quartic is of the first species, the first thirteen characteristics can be found from equations (10) to (15) of §107 by putting if = iV = 2. For the second species, let the suffix 1 refer to the quartic; then Wi = 4, via = 2, if = 3, iV=2; also the 102 TWISTED CtTRVES AND DETELOPABLES two straight lines form an improper conic having one apparent node; hence /i2=l- Substituting in (23) of §109 Ave obtain h^ = 3, The remainder of the thirteen characteristics can be obtained from (4) and (5) of § 104 by putting w = 4, h = S, S = K = T = 'U7 = 0, and t = 0, 1 or 2, The last eight can be obtained in both cases from the Salmon- Cremona equations. Quartics of the First Species. 138. Through every quartic of the first species four, three or two qiiadric cones can be drawn according as the curve is anauto- tomic, nodal or cuspidal. Let S, S' be the two quadrics containing the curve, then the equation of any other quadric passing through it is S + \S' = 0, and the condition that this should be a cone is that its dis- criminant should vanish, which furnishes a quartic equation for determining X. When the quartic is nodal, let A be the vertex of one of the cones, B the node and ABD the tangent planes to both quadrics at B ; then their equations may be written S = aa^ + C72 -1- dS' + 2//37 + 2g'ya + 2locS +2nyS = j and the discriminant of S + \S' is (f+\y(P-ad-a\) = (2), which shows that two of the cones coincide. To find the condition that the quadrics should have stationary contact at B, eliminate y from (1) and we obtain aoL^ + cS'/4>l3' + {d-f)S' + 2la8 - (ga + nS) S7/3 = 0. This is the equation of a quartic cone, whose vertex is G, which stands on the curve, and the condition that the coefficient of /3^ should be a perfect square is l^ = a (d —f), which reduces (2) to a (f+ xy = 0, and shows that three of the cones coincide. 139. If a plane passing through two fixed points on the quartic intersects the curve in two other points P and Q, the line PQ envelopes a quadric which contains the quartic ; also four planes of the system touch the curve. Let B and C be the two fixed points; A and D the vertices QUARTICS OF THE FIRST SPECIES 103 of two of the quadric cones which contain the quartic ; then its equations may be expressed in the form g2 + 2/37 = Oj ■(3). pa? + 2j9'/37 + 2^ 7a + ^r a/3 = j The equation of any plane through BG is S = Xol, whence the chord PQ is the intersection of this plane and the plane {p-p'\^)oL + ^'r^^-'lr'^ = (4), and the envelope of this line is obtained by eliminating A, and is pa? - p'B^ + 2q'ya + 2r'a/3 - 0, which is the result of eliminating ^y between (3). The condition that the plane B = Xa should touch the quartic is that the cone Va^ + 2yS7 = 0, and the second of (3) should touch ; which by eliminating y8 can be shown to be (p-pX'f = 8q'r'X', and furnishes a quartic equation for determining \. 140. When the vertex of the cone standing on the quartic lies on the curve, the cone will be a cubic cone which is anauto- tomic, nodal or cuspidal according as the quartic belongs to one or other of these species ; for since a straight line cannot cut a quadric surface in more than two points, the cone cannot have any apparent nodal generators. For the same reason the quartic cannot have any trisecants. Also any stationary tangent plane to the cone is an osculating plane to the quartic ; and since anauto- tomic cubic curves possess 9 points of inflexion, it follows that 9 osculating planes can be drawn to the quartic through any point on the curve. Again let P, Q, R be any three collinear points of inflexion on a plane section, and let the generators OP, OQ, OR cut the quartic in p, q and r ; then p, q and r form a triplet of points which possess the property of lying in the same plane, the osculating planes at which pass through a point on the curve. Moreover since a real straight line can be drawn through the three real points of inflexion, and also through each real and two conjugate imaginary points of inflexion, there are altogether four triplets corresponding to a point on the curve. 141. These results can be generalized. Let be any point in space ; then since an anautotomic twisted quartic curve 104 TWISTED CURVES AND DEVELOPABLES possesses two apparent nodes, the projection of the curve on a plane is a plane binodal quartic curve ; and since such a curve possesses 12 points of inflexion, it follows that through any point 0, 12 osculating planes can be drawn to the curve ; in other words, the curve is of the 12th class as we have already shown by means of the Pliicker-Cayley equations. By a known theorem*, the 12 points of inflexion of a plane binodal quartic will lie on a cubic curve, provided the four points in which the nodal tangents intersect the curve are collinear ; if therefore the point be chosen so that the four points in which the apparent nodal tangent planes intersect the curve lie in a plane passing through 0, the points of contact of the 12 osculating planes passing through will lie on a cubic cone. When the twisted quartic possesses an actual node, the projection of the curve will be a trinodal quartic, in which case only 6 osculating planes can be drawn through 0, and their points of contact lie on a quadric cone which passes through two generators OS, OS' of the quartic cone, which correspond to the S points of a plane trinodal quartic curve. Also the theorems of plane trinodal quartics relating to the conies which pass through (i) the points where the nodal tangents intersect the curve and (ii) the points where the tangents from the nodes touch the curve can be adapted in like manner to nodal twisted quarticsf . 142. When the excentricity of an ellipse is equal to (\/5 — l)/2\/2, the circles of curvature at the extremities of the minor axis intersect in two points E, E', which respectively lie on the circles of curvature at the extremities of the major axis ; and the inverse of the ellipse with respect to one of these points is a trinodal quartic having 3 points of undulation. Now the four o- planes of a nodal twisted quartic form a tetrahedron, and the cone standing on the curve whose vertex is any one of the vertices of the tetrahedron is a trinodal quartic cone of this character. The developables enveloped by the osculating planes to the three kinds of quartics of the first species have been discussed by CayleyJ. * Richmond, Proc. Lond. Math. Soc. vol, xxxiii. p. 218 ; Basset, Quart. Jour. vol. XXXVI. p. 44. t Basset, American Journal, vol. xxvi. p. 169. See also Appendix I. X C. M. P. vol. I. p. 486 ; Cmnb. and Dublin Math. Jour. vol. v. p. 46. The following papers relate to these curves. Hesse, Crelle, vol. xxvi. ; Reye, Ibid. QUARTICS OF THE SECOND SPECIES 105 143. A quartic of the first species is the partial intersection of a quadric and a cubic which possess a common conic. Let 8 = 0, 8=0 be the equations of the conic ; then the equations of any cubic and quadric surface which contain this conic are BX = Su, Sv = S (5), where 2 is another quadric, and u, v are planes. Eliminating S and S we obtain 'Z = uv (6), which shows that the quartic is the complete intersection of (6) and the second of (5). This theorem is true when the conic degrades into a pair of intersecting straight lines. Quartics of the Second Species. 144. A quartic of the second species is the partial intersection of a quadric and a cubic surface possessing a line in common, which is an ordinary line on the quadric and a nodal line on the cubic. Let a quadric and a cubic intersect in the lines CD and (w, v) ; then their equations may be taken to be 0Lu = ^v, a{uU + vV) = ^{uU'+vV') (1), whence eliminating {u, v) we obtain an equation of the form Pa^ + 2Qa^ + R^^ = (2), where P, Q, R are planes. The quartic is therefore the partial intersection of (2) and the first of (1), which proves the theorem. 145. A quartic of the second species cannot have any actual double points. Since CD is an ordinary line on the quadric and a nodal line on the cubic, we may suppose that the two surfaces touch at A , in which case A will be a node on the quartic. Hence if ABC be the common tangent plane to the two surfaces, we must have u=p0+r8, v=p2 + F'/3 + Q'j + R'8. Also if P = 0^ + Hy + KS, Q=fa+g/3 + hy + k8, vol. c. ; and Annali di Mat. vol. ii. ; Chasles, Comptes Rend. vols. Lii., liv. ; Zeuthen, Acta Math. vol. xii. ; Schrotter's treatise on the Theorie der Raumcurven ^ter Ordnung, Leipzig, 1890. 106 TWISTED CUBVES AND DEVELOP ABLES the conditions that 8 should be the tangent plane to (2) at A are G + 2f=0, H=0, which reduce (2) and the first of (1) to the forms kx^B + 2a^ (g^ + hj + kB) + R^' = 0, raB = ^{P'^+Q'y + R'B), and show that the line (/3, 8) or AG lies in the quadric and cubic. Hence the two surfaces intersect in three straight lines and a residual twisted cubic curve. 146. The developable D which is the envelope of the plane {a,b,c,d,e^ d,iy^O (3) has been discussed by Cay ley* and various other writers ; and we shall show that it is the reciprocal polar of a curve which includes all nodal and cuspidal quartics of the first species and all quartics of the second species. Its characteristics are v = 6, n = 6, m = 4, /c = 4, a; = 4, y = 6, h = 6, g = S, and therefore those of the reciprocal polar E' of the developable are y = 6, m = 6, w = 4, a = ^, a? = 6, y = ^, h = S, g = Q, which are those of a quartic curve of the second species and first kind. Writing as usual I = ae- ^hd + 3c .(4), /= ace + 26ccZ — ad"^ — Ife — c^ ] it follows that the equation of D, which is the discriminant of (3), is I^=27J' (5), * "On the developable derived from an equation of the fifth order," C. M. P. vol. I. p. 500; Camb. and Dublin Math. Jour. vol. v. p. 152. In this paper the discriminant of a binary quintic is given in a form which would repay a geometrical examination. " On certain developable surfaces," C. M. P. vol. v. p. 267 ; Quart. Jour. vol. VI. p. 108. "On the reciprocation of a certain quartic developable," C. M. P. vol. v. p. 505; Quart. Jour. vol. vii. p. 87. "On a special sextic developable," C. M. P. vol. v. p. 511 ; Quart. Jour. vol. vii. p. 105. The conditions for equalities amongst the roots of a quintic equation have been discussed by Sylvester, Phil. Trans. 1864, Collected Papers, vol. ii. p. 452 ; and these results have important applications with reference to the developables and curves derived from the binary quintic (a, b, c, d, e, f\d, 1)'' = 0. QUARTICS OF THE SECOND SPECIES 107 and those of E, which are the conditions that (3) should have three equal roots, are /=0, J=0 (6), and the four cusps, which are the conditions that (3) should have four equal roots, are determined by the equations a/b = h/c = c/d = d/e (7). The nodal curve on D is found from the conditions that (3) should have a pair of equal roots, and its equations are a

8e' + 3Xa^- - 2X^6 + X/3 = 0. . .(17), and the node is at the point D. 150. The tangents at the points of contact of a doubly osculating plane intersect at a point, which is a node on the nodal curve. By (8) and (14) the equations of the nodal curve are a' = X^\ 2/8^ + aS^ = S^yS, which represent a pair of conies whose planes intersect in the line GD, which does not form part of the nodal curve. Also since both conies intersect in the point C, and nowhere else, (7 is a node on the nodal curve ; and this is the point where the tangents AG and BG, at the points of contact of the doubly osculating plane, intersect. This theorem is a general one. 151. Guspidal quartics and their reciprocals. We have shown in § 124 that if in (3) we put 6 = 0, the curve E will possess a stationary tangent plane. Let it be ABG ; A its point of contact ; then we may take a = S, 6 = 0, c = l3, d = y, e = a ....(18), 110 TWISTED CURVES AND DEVELOP ABLES and the equation for D becomes a?h^ + 9a2/9S + Tia^' = 27 (a^ - 7^) {(a/3 - 7^) g _ 2^8^} . . .(19), which shows that any plane section of (19) through A has a tacnode thereat, hence : The points of contact of the stationary planes are tacnodal points on E. This theorem is a general one. 152. Equations (6), which determine the edge of regression, now become aS + 3yS2 = 0, a/38-V-/8^ = (20), from which we deduce 4a/3-37^ = (21), which shows that the curve is the complete intersection of (21) and the first of (20). Accordingly the curve is a cuspidal quartic of the first species, which possesses one cusp and one stationary plane, and is therefore its own reciprocal. Hence : A cuspidal quartic is the edge of regression of the developable enveloped by the plane 86^ + 6^6' + 47^ + a = 0, and the parametric equations for the coordinates are a = S6\ ^ = d\ y = -se', S = -l. 153. Quartics of the second species having points of inflexion. We have shown in § 124 (iv) that the condition for such a point is c = \a; and if we put m = 4, t = 1, o- = w = t = in (16) of § 127, it will be found that the characteristics of E are the reciprocals of those of a quartic curve of the second species which has one point of inflexion. Hence such curves are the reciprocal polars of the developables enveloped by the plane (a, b, Xa, d, e'^O, 1)^ = (22). 154. A quartic curve which has two points of inflexion is the reciprocal polar of the developable (oi8-4^yy^27{(xy' + fi''Sy (23). Equation (23) is the discriminant of (22), when X = and a = 8, b = y, d = 0, e — a, and it may be established as follows. Let A be one point of in- flexion, D the other; also let AB be the tangent at A, DC that at D. Then we may take a = B, b=py + q8, c = 0, d = roL + s^, e — a, QUINTIC CURVES 111 and the discriminant of (22) becomes {aS - 4 (p7 + qS) (ra + s^)Y = 27 (S (ra + sjSy +oi(py + qSfY (24). The form of (24) shows that AB and CD are double lines on the developable ; also the term involving the highest power of /3 is 27s^yS^S2, and since B may be any point on AB, it follows that the line is cuspidal. In like manner GD is a cuspidal line. The plane ra + s^ may be any plane through GB, let us therefore choose it for the plane ^ ; then r = and we may take s = 1 ; hence the term involving the highest power of a is a°S^ The point A is now one where the stationary tangent touches the curve, and is therefore a cubic node of the sixth kind on the developable D. In like manner if D is the point of contact of the other stationary tangent, p — 1, q=0, and (24) becomes (aB-4>/3yy = 27(aY' + ^Sy (25). 155. It thus appears that anautotomic quartic curves of the first species constitute a class of curves sui generis; but that nodal and cuspidal quartics, and also all quartics of the second species, constitute a class of curves which possess many features in common. In particular they are all unicursal curves, and are also included amongst those which are the reciprocal polars of the developables enveloped by (3). No quartic of the second species can possess a double tangent, since the latter would be a line lying in the cubic and quadric surfaces of which the quartic is the partial intersection ; in which case the quartic would degrade into the double tangent and a cubic. A historical account of unicursal quartic curves, together with a list of memoirs, has been given by Mr Richmond, in Trans. Gamh. Phil. Soc. vol. xix. p. 132. Quintic Gurves. 156. There are four primary species of twisted quintic curves *. I. Quintics which are the partial intersection of a quadric and a cubic surface, the residual intersection being a common straight line. These possess four apparent nodes and may also have two actual double points, which may be nodes or cusps. * Cayley, C. M. P. vol. v. p. 15. 112 TWISTED CURVES AND DEVELOP ABLES II. Quintics which are the partial intersection of two cubic surfaces, the residual intersection being a quartic curve of the second species. These have five apparent nodes, and may also possess an actual double point. III. Quintics which are the partial intersection of two cubic surfaces, the residual intersection being a twisted cubic curve and a straight line. These have six apparent nodes. IV. Quintics which are the partial intersection of a quadric and a quartic surface, the residual intersection being three generators of the quadric belonging to the same system. The number of apparent nodes is obtained from the equation 2 {h - h') = (n- n') {M-\){N-\ ), where the unaccented and accented letters refer to the quintic and the residual curve. In the four respective cases h' = 0, 3, 4, 3 ; w' = 1, 4, 4, 3 ; which gives h = 4i, 5, 6, 6. Since the cone standing on a twisted quintic curve is a quintic cone having at least four double generators, a great many properties of such curves may be derived from those of plane quintic curves, which have been discussed by myself* ; I shall therefore briefly consider the four species. 157. First Species. If U, V are quadric surfaces, the simplest form of the equations of curves of this species is U, a, ^ V, y, B from which it can be shown as in § 102 that the quintic is also the partial intersection of two cubic surfaces, whose residual intersection is a quartic of the first species. We shall now show that : 158. A quintic of the first species is the partial intersection of a quadric and a quartic surface, the residual intersection being a twisted cubic. Let U, F, W be quadric surfaces \ p, q, r constants ; u, u', &c. planes ; also let \ = vw — v'w, fx = wu' — w'u, V = uv — u'v, * Quart. Jour. vol. xxxvii. pp. 106 and 199. See also, "On plane quintic curves •with four cusps," Rend. Palermo, vol. xxvi. p. 1. = (1), QUINTIC CURVES 113 and consider the equations p\ + qfjb + rv = oi (2) ; uX + Vfi -\- wv = OJ the first two equations represent a quartic and a quadric surface which intersect in the twisted cubic (X, fi, v) and in a residual quintic curve, whilst the last one is an identity. Eliminating (X,, fi, v) we obtain U(rv — qw)+ V(pw — ru)+ W {qu-pv) = (3), whilst the second of (2) may be written u' (rv — qw) + v' (pw — ru) + w' {qu — pv) = (4). Equations (3) and (4) represent a cubic and a quadric which both contain the residual quintic, and consequently the latter is of the first species. 159. Second Species. Let U=a^-yS, V={a^+by)oL + (c^ + dy)B (5), where (a, b, c, d) are arbitrary planes ; then U =0, V=0 represent a quadric and a cubic surface which intersect in the lines BC and AD ; hence the residual intersection is a quartic of the second species. From (5) eliminate successively {^, y) and (a, B) and we obtain by' + (a + d)^y + c^' = Oj which represent a pair of cubic surfaces on which BG and AD are nodal lines respectively. These intersect in a quartic of the second species and a residual quintic curve of the same species. 160. Third Species. The equations of these quintics may be expressed by means of the system of determinants p, s, P, S q, t, Q, T r, u, M, U where the small letters represent arbitrary planes; whilst the capital letters represent six planes^ passing through the same straight line but otherwise arbitrary. For if \ — qu — rt, /j, = rs — pu, v = pt — qs, B. 8 .(6), = (7), 114 TWISTED CURVES AND DEVBLOPABLES the determinants are equivalent to 8k + Tfi+Uv = 0, and these are the equations of two cubic surfaces each passing through the twisted cubic (X, /j,, v) and the common line of intersection of the six planes. The residual curve is therefore a quintic. 161. Fourth Species. The lines CD and AB are generators of the quadric 07 = ySS, and the equations of any other generator of the same system are a = Xh, ^ = Xy; and the equation of any qnartic containing the curve may be taken to be P(a-Xh) + Qi\y-^) = (8), where P and Q are quaternary cubics, which have to be determined so that (8) vanishes when a = 0, /3 = ; or when 7 = 0, S = 0. Let n = aa^ + ba/3 + c^\ a' = ^ 7^ + 57S + Gh\ where a, A ... are constants ; then the values of P and Q may be written P = /80 + 7II' + a (a^i + /3o-i + v.^ + ^ (awj + jSr^ + lu^), Q = an + m' + a (a< + /Scr/ + <) + /3 (aw,' + ^t^ + w.^\ where the suffixed letters denote quantics of (7, S). Denoting the last two terms by f/, U', (8) becomes (a7 - ySS) (\a + a') + {a-\t)U+ (\7 -^)U' = 0, which shows that the curve is the intersection of the quadric ay = ^B and the quartic (a-XS) U+{\y-^) U' = (9). By means of the equation of the quadric, (9) may be reduced to ^3a+ 233/8 + ®4 = 0, where the old English letters denote binary quantics of (7^ 8), hence : The curve is the partial intersection of a quartic which has a triple line, and a quadric which passes through the line. The following papers* relate to quintic curves; and the con- sideration of sextic curves will be postponed until we discuss the Theory of Residuation. * Bertini, Collect. Math. 1881; Berzolari, Lincei, 1893; Weyl, Wiener Berichte, 1884-5-6; Montesano, Ace. Napoli, 1888. CHAPTER ly COMPOUND SINGULAEITIES OF PLANE CURVES 162. Although the geometry of surfaces is the object of this treatise, yet the theory of their singularities cannot be properly understood without a more detailed account of the corresponding portion of the theory of plane curves, than is contained in my treatise on Cubic and Quartic Curves. I shall therefore devote the present chapter to the consideration of the compound singularities of plane curves*. 163. Pllicker's equations show that the simple singularities of a curve are four in number, viz. the node, the cusp, and their reciprocals the double and the stationary tangent ; and also that every algebraic curve possesses a determinate number of these singularities which can be calculated from the formulae he gave. From this it follows that every other singularity, which an algebraic curve can possess, is a compound singularity formed by the union of two or more simple singularities. Compound singularities may be divided into three primary species. First, point singularities, which are exclusively composed of nodes and cusps. Secondly, line singularities, which are ex- clusively composed of double and stationary tangents. Thirdly, mixed singularities, which are composed of a combination of point and line singularities. 164. The point constituents of a singularity can be determined in the following manner. Pliicker's first equation is 2B + SK = n(n-l)-m (1), where 8 and k are the number of constituent nodes and cusps, and 2S + Sk is the reduction of class produced by the singularity ; and since the degree n of the surface is given, it follows that as soon * Basset, Quart. Join: vols, xxxvi. p. 359, xxxvii. p. 313. 8—2 116 COMPOUND SINGULARITIES OF PLANE CURVES as its class m has been ascertained, (1) furnishes one relation between the unknown quantities S and k. Another equation exists of the form Z + K = \ (2), where X is the number of constituent double points ; and as soon as X has been found, (1) and (2) furnish two equations for deter- mining 8 and k. The line constituents can usually be found by forming the reciprocal singularity, and ascertaining the number of its con- stituent nodes and cusps. The only point singularities which exist are multiple points of order p, the tangents at which have (^-l-l)-tactic contact with the curve at the point. If any tangent has a higher contact, the singularity is a mixed one. 165. If r tangents at a multiple point of order p coincide, its constituents are B = ^p(p — l)-r + l, K = r — 1. Since the properties of a multiple point of this kind are the same on a curve of degree p + 1 as on one of higher degree, we may employ the curve aYUp_r+Up+i = (3), the triangle of reference being chosen so that A is the multiple point, and AB the line which coincides with the r coincident tangents. The first polar of C, which may be any arbitrary point, is aY~^ {rup-r + yu'p-r) + u'p+i = (4), where the accents denote differentiation with respect to 7. Eliminating a between (3) and (4) we obtain Y~^ {yUp-rU'p+i — (rUp^r + yu'p-r) Up+i] = 0, which shows that the first polar of G intersects the curve in 2p —r+ 1 ordinary points ; hence m = 2^ — 7' 4- 1, and since the degree of the curve is p 4- 1, we obtain from (1) 28 + SK=p(p-l) + r-l (5). Since the point G is arbitrary, it follows that if the curve has another double point we may suppose it situated at G, in which MULTIPLE POINTS 117 case the terms in 7^+^ and 7^ must be absent, and (3) reduces to the improper curve showing that the deficiency of (3) is zero ; whence h + K = ^p{p-l) (6). Solving (5) and (6) we obtain the required result. When all the tangents are distinct, r =1, and the constituents of the point are S = ^jp(|) — 1), a; = 0. It can also be shown that if r tangents coincide with a particular line AP, and s tangents with another line A Q, the constituents of the point are h = ^p{p-l)-r — s + % K=r+s — 2. It is impossible for a multiple point to be composed exclusively of cusps, for if all the tangents coincide r = p, and the constituents are S = i(P-l)(p-2), fc^p-l. 166. Reciprocating the theorem of § 165, we obtain : If a multiple tangent of order p has (r + V)-tactic contact at one point, and hitactic contact at p — r points, its constituents are '^=ii'(i'-l)-^+l. t, = r-l (7). 167. Let r tangents at a multiple point of order p coincide ; then if t he the number of tangents which can he drawn from the point, and m the class of the curve t = m — 2p + r—l (8). The reciprocal polar of the multiple point is a multiple tangent to the reciprocal curve, whose degree is m. The tano-ent has (r + l)-tactic contact at one point, bitactic contact at p—r points, and intersects the curve at t ordinary points ; hence t + 2 (p — r) + r + 1 = m, giving t = m — 2p + r — l, and the number of ordinary points of intersection are the reciprocal polars of the tangents drawn from the multiple point on the original curve. When all the tangents are distinct r = 1 and t = m-2p (9). 118 COMPOUND SINGULARITIES OF PLANE CURVES In the same way it can be shown that if r tangents coincide with a line AP and s with a line AQ, the value of t is t = m -1p + {r -I) + {s -1) (9 a). 168. When the number of constituent double points in a singularity is unequal to ^p (p — 1), the latter cannot be a multiple point but must be a mixed singularity. It is also possible for a singularity to possess this number of double points without being a multiple point. Thus the point constituents of an oscnode are S = 3 ; and the distinction between a triple point of the first kind and an oscnode is that (i) the three nodes move up to coincidence in an arbitrary manner, whereas in an oscnode they move up to coincidence along a continuous curve ; (ii) the triple point has no line constituents, whereas those of an oscnode are t = 3. 169. If an arbitrary straight line through a point P, which is not a multiple point of order p, intersects the curve in p coincident points at P, then P is called a singular point of order p. The rhamphoid cusp and the oscnode are examples of singular points of order 2. Also if from an arbitrary point on a tangent, which is not a multiple tangent of order p, m—p tangents can be drawn to a curve of class m, the tangent is called a singular tangent of order p. The distinction between multiple points and singular points is of importance in the theory of compound singularities, 170. The theorem of § 24 is applicable to plane curves, and affords a ready means of determining the number of constituent point singularities. It is : If a node moves up to coincidence with a multiple point of order p along the line AB, the equation of the curve is OL^-^rfUp_^ + OL^'-^-^r^Up + tt'^-^-^M^+s + . . . M„ = . . .(10). The equation of a curve having an ordinary multiple point of order p at u4 is a^-fvp + a'^-^-iv^+i + . . . Wn = (11). If the curve has a node at a point P on J.5,the line AB must have p-tactic contact with the curve at A and bitactic contact at P; hence when P coincides with A, the line AB must have (p + 2)-tactic contact at A. Similarly the first polar of G, which is any arbitrary point, must have ^-tactic contact at A. These conditions reduce (11) to (10), and the point constituents of the singularity are h = ^p {p) — \) -{■ \. TACNODAL BRANCHES 119 171. To find the line constituents, we must consider the reciprocal singularity, and for this purpose we may employ the curve a^y^Up-^ + ay Up +Up+2 = (12). From § 170 it follows that m = 4p, also 2p tangents can be drawn from A to the curve ; hence the reciprocal singularity is a tangent to a curve of degree 4ip, which touches it at ^ — 2 distinct points, corresponding to the distinct nodal tangents Up^^ = ; also the tangent intersects the curve in 2p points, corresponding to the 2p tangents drawn from A ; and it touches it at \- 4 coincident points at a point A', which is the reciprocal of the tangent AB to the original curve. If we write down the first polar of (12) with respect to B, which may be any arbitrary point on AB, and eliminate a7, the result is a binary quantic of (yS, j) of degree 4>p — 2, which shows that 4p — 2 tangents can be drawn to (12) from an arbitrary point on AB. Hence an arbitrary line through ^1' cuts the reciprocal curve in 4p — 2 ordinary points, and therefore A' is a singular point of the second order. The reciprocal singularity is therefore a tacnodal tangent, which has bitactic contact with the reciprocal curve at p — 2 points, and its constituents are 8 = 2, r=ip(p-l) + l, whilst the original singularity is a multiple point having one pair of tacnodal and p -2 ordinary branches, and its constituents are B = ^p(p-1) + 1, T = 2. 172. The above results are true when there are any number of tacnodal branches, and may be generalized as follows : (i) // a multiple point of order p has s pairs of tacnodal branches and p — 2s distinct ordinary branches, its constituents are 8 = ^p(p-l) + s, T=2s. Putting p = 2s, it follows that (ii) If a rnidtiple point of order 2s has s pairs of tacnodal and no ordinary branches, its constituents are h = 2s\ T = 2s. 120 COMPOUND SINGULARITIES OF PLANE CURVES The reciprocals of these singularities are : — (iii) A multiple tangent which touches a curve at s tacnodes and has bitactic contact with the curve at p — 2s points ; and its constituents are 8 = 2s, T = i^ (p - 1) + s. (iv) A multiple tangent which touches the curve at s tacnodes and nowhere else ; and its constituents are 8 = 2s, T = 2s2. 173. We must now consider how these results are modified when some of the branches coincide ; and we shall show that every ordinary branch which coincides with a tacnodal branch changes a node into a cusp, whilst every pair of tacnodal branches which coincides with another pair of tacnodal branches changes two nodes into two cusps. The first theorem is as follows : — If a midtiple point of order p consists (i) of one pair of tacnodal branches, (ii) of r ordinary branches which coincide with the pair of tacnodal branches, (iii) of p — r —2 distinct ordinary branches ; its constituents are h = ^p{p—l) — r-\-l, K = r, T = 2. The curve a" (X/3 + fjiyy+Hp-r-2 + oLUp+i + Up+2 = (13) has a multiple point at A consisting of p— r— 2 distinct and r + 2 coincident branches ; and if an additional double point moves up to coincidence with A along AB, it can be shown as in § 170 that (13) becomes 0?Y^^Up_r-2 + OSiUp + Wp+2 = (14). Write down the first polar of G, which may be any arbitrary point, and eliminate a, and the result will be a binary quantic of {B, 7) of degree 4p — r. Whence 2a + 3/c = (j9 + 2) (p + 1) - 4^ + r =^(^-l)+r + 2. Also 8 + /c = ^p (p— 1) + 1, whence h = {p{p — l) — r+\, K = r (15), which give the point constituents of the singularity. The reciprocal singularity consists of a multiple tangent which has bitactic contact with the reciprocal curve at p — r — 2 points and MULTIPLE TANGENTS 121 touches it at q points at a tacnode, and also cuts the curve at 2p points, which correspond to the 2p tangents which can be drawn to (14) from A. Hence 4^ _ r = g + 2 (p - r — 2) + 2^, which gives g = r + 4. Accordingly the line constituents of the original singularity are T = 2 ; and the reciprocal singularity is : — (i) A multiple tangent which touches the curve at r+ 4! points at a tacnode and has bitactic contact with it at p — r — 2 distinct points ; and its constituents are 8=2, T = lp(p — l)—r + l, L = r. Also the coincidence of each successive ordinary point increases the contact by 1, and converts a double tangent into a stationary one. Let p — r — 2=0, then : — (ii) The constituents of a tacnodal tangent which touches the curve at ?' + 4 points at a tacnode and nowhere else, are B = 2, T = ir(r+l) + 2, t = n Also each additional point of contact after the (r+4)^A adds one stationary and r + 1 double tangents to the constituents of the singularity. 174. The theory of coincident tacnodal branches is contained in the following theorem : — // a multiple point of order p has s pairs of tacnodal branches of which r pairs are coincident, r> 1, and p — 2s ordinary branches; its constituents are* h = ^p{p-l)-\- s-2r + 2, K=2r-2, T = 2s-r+l. This theorem, so far as its point constituents are concerned, may be proved by the previous methods ; but the portion relating to the line singularities will be proved in the next section. We notice the following special cases. If there are no ordinary branches p = 2s, whence (i) If a midtiple point of order 2s consists of r pairs of coincident and s — r pairs of distinct tacnodal branches, its con- stituents are S=2s2-2r+2, K = 2r-2, r = 2s-r + l. * When r=l, all the tacnodal branches are distinct. 122 COMPOUND SINGULARITIES OF PLANE CURVES If all the tacnodal branches coincide, s = r, whence (ii) If a multiple point of order 2s consists of s pairs of coincident tacnodal and no ordinary branches ; its constituents are S=2s^-2s + 2, K=2s-2, t = s + 1. (iii) If all the tacnodal branches coincide, and there are p — 2s ordinary branches, the constituents of the multiple point are h = \p{p-l)-s+2, K = 2s-2, T=s + 1. 175. We must now examine the reciprocal singularity. Con- sider the two curves a?-l) + s-2r +2, / = 2r-2. BIRATIONAL TRANSFORMATION 123 (^) A tangent touches a curve (i) at s — r distinct tacnodes ; (ii) at a point composed of the union q/ r + 1 collinear nodes, and at no ordinary points ; its constituents are a = 2s-r + l, T=2s2-2r + 2, i = 2r-2. (7) A tangent touches a curve at a point composed of s + 1 coincident collinear nodes; its constituents are g = s + l, T = 2s2-2s + 2, t = 25-2. (S) A tangent touches a curve at a point composed of s-\-l coincident collinear nodes and at p — 2s ordinary points ; its constituents are S = s+1, T = \p{p-l)-s + 2, i. = 2s-2. Birational Transformation. 176. We shall now explain the theory of birational trans- formation, and shall show how it may be employe^ to investigate the constituents of the compound singularities of curves. The conic a^ = /3y touches the sides AB, AG of the triangle of reference at B and C. Let P be any point (^, tj, ^) ; and let AP cut the polar of P with respect to this conic in a point P', whose coordinates are (f, r]', ^'). The polar of P is 2a^-^^- 777 = 0, and since this passes through P', we have m'-^v-vK' = o '. (1). But the equation of AP' is, W = 7/r = ^^ (say), whence tj/t}' = ^/^' = k (2). Substituting in (1) we obtain Accordingly from (2), we have which is the equation connecting the coordinates of P and P'. It follows from the above construction, that any point on BC except B and G corresponds to A ; any point on ^P except A 124 COMPOUND SINGULARITIES OF PLANE CURVES corresponds to B; and any point on AG except A corresponds to a 177. A node which does not lie on the sides of the triangle ABG transforms into a node. The curve u^U + uvV+v^W=0, where {u, v) are straight lines, and U, V, W are ternary quantics of degree n — 2, has a node at the point of intersection of the lines u = 0, v = 0; and if this curve be transformed by means of (3), 21 and V will become conies circumscribing the triangle ABC and intersecting in a fourth point P' which corresponds to the node {u, v). The point P' is obviously a node; and the theorem can be extended to multiple points of any order. 178. Let a curve cut BC in two ordinary points P and Q, which can always be effected by making B and C multiple points ; then the transformed curve will have a node at A. And generally, if the curve cut BG in s ordinary points, the transformed curve will have a multiple point of order s at A ; also since any pair of ordinary points gives rise to a node at A, it follows that the number of constituent nodes of a multiple point of order s is equal to the number of combinations of s things taken two at a time, that is, to ^s (s— 1). The directions of the nodal tangents at A are determined as follows. Let there be two ordinary points P and Q on BG ; and let p, q be two points on the curve in the neighbourhood of P and Q. Then if p', q' be the corresponding points, Ap' and Aq', and ultimately AP and AQ, will be the directions of the nodal tangents at A. Hence if P and Q coincide, AP and AQ will also coincide ; accordingly, if the curve touches BG at P and does not intersect it at any ordinary points, the transformed curve will have a cusp at A. If BG touches the curve at P and intersects it at one ordinary point Q, the transformed curve will have a triple point of the second kind at A, consisting of a cusp and a branch through it ; and its constituents are 8=2, k = 1. And generally if the curve touches BG in r coincident points and intersects it in s — r points, the transformed curve will have a multiple point of order s a.t A at which r tangents coincide; and the constituents of such a point are S = ^s(.9— 1) — r+ 1, K = r — 1. BIRATIONAL TRANSFORMATION 125 If BG cuts a curve at a node and in no ordinary points, the transformed curve has a tacnode at A ; hence each of the two coinci- dent points of which the node is composed transforms into a node, whilst the two branches which pass through the node transform into the two branches which touch one another at the tacnode. For example the equation a?Ui + ct^yVjU^ + ^^y%^ = represents a sextic having nodes at B and 0, and a third node at the point a = 0, Vj = ; and this transforms into the curve a.%^ + a.ViU2 + W4 = 0, which is a quartic having a tacnode at A. If BG cuts the curve at a node and p — 2 ordinary points, it can be shown in the same way that the transformed curve has a multiple point of order p a,t A consisting of one pair of tacnodal and p — 2 distinct ordinary branches. And since its point con- stituents have been shown to he 8 = ^ p (p — 1) + 1, it follows that each of the two points which coincide at the node gives rise to a node, whilst every ordinary point in combination with either of the nodal points or with another ordinary point gives rise to a node. Also the theorem of § 172 (i) shows that this is true for any number of nodes and ordinary points on BG; and it follows from § 173 that if r ordinary points on BG coincide with a node, the effect is to convert r of the constituent nodes of the trans- formed singularity into cusps. 179. Before considering the case of a cusp, it will be useful to state that the equations of a quartic curve which has a tacnode, a rhamphoid cusp, an oscnode and a tacnode cusp at A may be written in the forms (oLUi + u^y +Ui = (4), {aui + U2y + UyUs = (5), (aui + Wg)^ + u^ {la. + m/3 -f- n '^ = 2, T = l, i = l. The reciprocal singularities are (iii) A rhamphoid cusp, whose cuspidal tangent touches the curve at p — 2 ordinary points ; and its constituents are 8=1, K = l, T = ^p(p-1), 1 = 1. (iv) A rhamphoid cusp, whose cuspidal tangent has quinque- tactic contact with the curve at the cusp, and touches the curve at jj — 3 ordinary points ; and its constituents are 8 = 1, K=l, T = ^p{p-\)-\, t = 2. BIRATIONAL TRANSFORMATION 127 Putting p=S,it follows that the constituents of a rhamphoid cusp whose tangent has quinquetactic contact with the curve are B = l, K = l, T = 2, 1 = 2. It can also be shown by birational transformation that the equation of a curve having a rhamphoid cusp at A, and AB as the cuspidal tangent, is a"-* (ilf/32 + Layf + a'^-^r ^h + a"^'7«*3 + ot^'-'u, + ... Un = . . .(9), and if the tangent has quinquetactic contact at A, M = and (9) becomes Za'^-272 + a'^-^ry^Ui + a^'-'^yus + a'^-^u, + . . . m„ = . . .(10). 182. If the line BO intersects a curve at an n-tuple point of the first hind and at no ordinary points, the transformed singularity consists of a 'niidtiple point formed by the union of two n-tuple points. Its constituents are 8=n{n — l), T= n{n — 1), and the equation of the curve is anyn + a^-iryu-i^^ + ^^^ ayu,n-2 + Um = 0. It will be sufficient to prove this theorem for a sextic curve, since the method of proof is the same for any other curve. The sextic curve a^Us + oi^Ui + 01.U5 + Mg = (11) has a triple point of the first kind at A, and if it has another triple point at a point P on AB, it follows that A and P are nodes on the first polar and ordinary points on the second polar of G, which may be any arbitrary point. Hence when A and P coincide, AB must have sextactic contact with the curve, quadritactic contact with the first polar, and bitactic contact with the second polar of G. This will be found to reduce (11) to the form a^yS _^ a^ry^u^ + OCjUi + Uq = (12). Hence the singularity is a singular point of the third order, the tanoent at which has sextactic contact with the curve. Writinsf (12) in the form a^Vi^ + a.^Vi^U2 + aViUi+UG = (13), and transforming birationally, we obtain a?UG + OL^jSryViU^ + al3^y^Vi^U2 + ^^y^Vj^ = (14), which is the equation of a curve of the 9th degree, having triple points at B and G and also at the point a = 0, Vj = on BG. 128 COMPOUND SINGULARTTIES OF PLANE CURVES The portion relating to the line constituents may be proved as follows. The reciprocal curve is of degree 18 ; also since the discrimi- nant of (12) is of degree 12, it follows that 12 tangents can be drawn from A to (12); hence if A' be the point on the reciprocal curve corresponding to AB, the tangent at A' has sextactic contact with the reciprocal curve at A\ The first polar of B, which may be any arbitrary point on AB, is a^y^U2+ap, the number of points absorbed at A is p{p + l)(q + l). (ii) But, if these are also lines of closest contact on the third surface, the number absorbed at A is p{p + l){q + 2). If Uq contain all the lines of closest contact, each will inter- sect (1) in only one ordinary point ; hence the number of points absorbed at ^ is (i? + l)Hg + 2)-(^ + l)(g + 2)-p(^ + l)=p(p+l)(^ + l). And if these are lines of closest contact on (1), they will not intersect (1) in any ordinary points; hence the number of points absorbed is {p + iy(q+2)-{p + l){q + 2)=p(p + l)(q + 2). The last two theorems are true* when p = q. 190. A multiple point of order p, the tangent cone at which is anautotomic, reduces the class'f by p {p — ly. When a surface has a multiple point of order ^ at J., the first polars of any two points have multiple points of orders p — 1 at A ; also if the nodal cone is anautotomic, this cone and the nodal cones at A to the first polars have no common generators ; hence A absorbs p{p—^y of the points of intersection of the surface and the first polars of any two points. Accordingly m = n(n— ly —p {p — 1)^ 191. Before finding the constituents of a multiple point, a few additional remarks on the compound singularities of plane curves will be necessary. If h nodes on a plane curve move up to coincidence in any manner whatever, the point constituents of the resulting compound * If a surface of degree ^ + 2 has a multiple point of order p ai A, the tangent cone from A obviously cannot possess any generators which have tritactic contact with the surface at some other point P ; hence the surface and its first and second polars with respect to A cannot intersect at any ordinary points, and therefore the number of points absorbed dA, Ais p (p + 1) (p + 2). A similar argument frequently gives a short cut to theorems of this character. t Segre, Ann. di Matem. Serie II. vol. xxv. p. 28. 134 SINGULARITIES OF SURFACES singularity are B nodes ; but it is otherwise in the case of cusps. For if K cusps move up to coincidence, it frequently happens that 2p of them are changed into 3p nodes, and this is especially the case when the cusps move up to coincidence along a continuous curve ; also since the reductions of class produced by a node and a cusp are respectively equal to 2 and 3, the class of the curve remains unaltered. The simplest example of the conversion of cusps into nodes is furnished by the oscnode. For the equation of a bicuspidal quartic curve can be expressed in the form S' = uv' (2), where u is the double tangent, v the line joining the cusps and ^ is a conic, which passes through the points of contact of the double tangent and has tritactic contact with the curve at each cusp; but when the line v touches the conic S, the two cusps coincide and the resulting singularity becomes an oscnode. If however the point constituents of an oscnode were two cusps, it would be possible for the quartic to have a third double point ; but if one be introduced, it can be shown in the following manner that the quartic will degrade into a pair of conies which osculate one another. Let ABC be the triangle of reference, A the oscnode, AB the oscnodal tangent ; then (2) becomes (ay + P^' + Q^ry + Ry^) = (la + m^ + ny)rf (3). Since G is an arbitrary point we may suppose it to be an additional node, the conditions for which are l = 2R, m = 2QR, n = R\ and (3) becomes {P^^ + QPy + OLyf + 2PR^'y' = 0, which represents a pair of conies. This shows that the union of the two cusps produces a compound singularity whose point con- stituents are three nodes ; and many other similar examples might be given. 192. The constituents of a multiple point of order p, the tangent cone at which is anautotomic, are* G^hpip-^h ^ = 0, * Basset, Bend, del Circolo Mat. di Palermo, vol. xxvi. p. 329. CONSTITUENTS OF A MULTIPLE POINT 135 where C and B are the number of constituent conic nodes and hinodes. Let A be the multiple point, D any point in space ; then since J. is a multiple point of orders ^ — 1 and p — 2 respectively on the first and second polars of the surface with respect to D, it follows that A absorbs p (j? — 1) (p — 2) of the points of intersection of the surface and its first and second polars with respect to D. Hence the number of distinct generators of the tangent cone from D, which have tritactic contact with the surface, and which are therefore cuspidal generators of the cone, is K = n{n-l){n-^)-p{p- 1) {p - 2) (4). Let V and ^l be the degree and class of the tangent cone from By then v=n{n—\.), /j, = n {n — iy—p(p — If (5) ; also let 8 be the number of distinct generators which are double tangents to the surface, and which are therefore nodal generators of the cone. Since the tangent cone at A is anautotomic, its class is p(p — 1), and therefore DA is a multiple generator of the tangent cone fi^om D of order p{p — 1), the tangent planes at which are distinct ; hence J. is a multiple point of the same character on the section of the cone by the plane ABC, and its point constituents are iP iP — ^){p^~P~ 1) nodes. Applying Pliicker's equations to the section of the tangent cone from D, we obtain fi = v{v-l)-p{'p-l){p'-p-l)-^h-^tc (6). Substituting the values of k, /ju and v from (4) and (5) in (6), we obtain B = ^n{n-l)(n-2){n-S)-^pip-l){p-2){p-S)...(7), hence if B' and k' are the number of nodal and cuspidal generators which are absorbed by the multiple point 8' = ^p{p-l)ip-2)(p-S), k'=p(p-1)(p-2)...{8). We shall now suppose that the multiple point at A is formed by the union of G conic nodes and B binodes. These double points are originally supposed to be isolated and to be arranged in any manner on the surface ; hence the tangent cone from D will possess two species of nodal and cuspidal generators, the first of which arises from the double points on the surface, whilst the 136 SINGULARITIES OF SURFACES second arises from generators which are double and stationary tangents to the surface. When the G conic nodes and B binodes coincide at A, all the generators of the first species and 8' + /c' of the secoud species will coincide with the line DA ; and we have to find the number of those of the first species. The multiple point at A on the section of the tangent cone from D is of order j9(p — 1), and is composed of double points of both species ; hence G-^B-\-h' + K' = ^p-'{p-\f-\p{p-\) (9), where h' is given by the first of (8), but nothing at present is supposed to be known about k except that it represents the effect of the coincidence of the p{p—^){p — 2) cuspidal generators of the second species. Also since the reduction of the class of the surface iQ p{p — \y, it follows that 2G+W=p{p-lf (10). Substituting the value of 8' from the first of (8) we obtain from (9) and (10) C=lp{p-i) (10^ - 19) - S/c'] •(11). 2/c'-5=3j?(p-l)(p-2) . Now if we supposed that the k distinct cuspidal genera- tors of the second species were equivalent after coincidence to p{p — l){p — 2) cusps, we should obtain from the last of (11) B = -pCp-l){p-2\ which is impossible, since B cannot be a negative quantity. This shows that the effect of coincidence is to convert the 2/c' cusps into f /c' nodes, which produces no alteration of the class of the tangent cone or of the surface, but makes G = lp{p-iy, B = 0. 193. When the tangent cone is autotomic, the investigation of the point constituents of any multiple point involves the solution of two distinct problems. In the first place the class m of the surface is determined by the equation m=n(n-iy-2C-W (12), and in the second place an equation exists of the form G + B = X (13). When the tangent cone is anautotomic the value of X by the preceding theorem is |p(j3 — l)^ and the theorem of § 24 usually CONSTITUENTS OF A MULTIPLE POINT 137 enables us to ascertain without much difficulty whether any change in the character of the multiple point is produced by the conversion of conic nodes into binodes, or by the union of additional double points with the multiple point. The principal difficulty is to determine the value of m, and we shall proceed to explain the methods by which this can be effected*. 194. When the nodal cone at a multiple point of order p has S nodal and k cuspidal generators, all of which are distinct, the reduction of class is p{p-iy+h-^2K, and the point constituents of the singularity are Let the equation of the surface be aup + Up+i = (14), then the first polars of C and D are au'p +u'p+i =0 (15), au"p + u"p+i = (16), where the single and double accents denote differentiation with respect to UpU p^i = Up^iU p \ -* "/• Equations (17) and (18) represent two cones of degree 2p, and their 4p^ common generators intersect the surface (14) at the points where it is intersected by (15) and (16) ; but these generators include the p{p + l) lines of closest contact, which do not give rise to ordinary points of intersection ; hence the number of the latter is reduced by p{p-{-l). Again, if we temporarily regard the cones Up, u'p and u"p as curves lying in the plane BCD, the last two will be the first polars of Up with respect to C and D; accordingly if AB is a nodal generator, it must be repeated once on the cones u'p and u'p, and twice if it is a cuspidal generator, but the three cones Up, u'p and u"p will not in general have any other common generator except AB. Hence every nodal generator on Up produces a further reduction * Basset, " Multiple points on Surfaces," Quart. Jour. vol. xxxix. p. 1. 138 SINGULAHITIES OF SURFACES in the number of common generators eqnal to 1, and every cuspidal generator reduces it by 2. Accordingly the number of ordinary points of intersection of (14), (15) and (16) is 4>p-—p(p+l) — 8— 2k, giving m = 4fp'^—p(p + l) — 8 — 2K = (p + l)p'' -p (p-iy-S- 2k, which shows that the reduction of class is given by the last three terms. We thus obtain 2G + SB=p(p-iy+B + 2K (19). From the theorem of § 24, it is easily seen that the reduction of class is not produced by the union of any additional double points with the multiple point; hence by § 193 G + B=lp{p-iy (20). Solving (19) and (20) we obtain the required result. 195. If the nodal cone at a multiple point of order p possesses a multiple generator of order q, such that r of the tangent 'planes are coincident, the constituents of the singularity are G=\p{p-\y-{q-\Y-r + \, B = (q-iy + r-l. (i) Let all the tangent planes along the multiple generator AB be distinct ; then since a multiple point of order q on a, curve gives rise to a multiple point of order q — 1 on the first polar, it follows that the first polars of two arbitrary points inter- sect in (q — ly coincident points at B ; hence if s be the additional reduction produced by the generator, s = (q — ly. (ii) Let r of the tangent planes along AB coincide ; then AB is a multiple point of order g- — 1 on the first polar, having 1 — 1 coincident tangent planes ; accordingly AB will be repeated r — 1 additional times on the first polars of two arbitrary points, so that s = (5' — l)^ + r— 1. This gives 2(7+ SB=p (p - ly + (q-iy + r-l, also G + B = ^p(p-iy, which proves the theorem. It does not appear to make any difference whether the pre- ceding compound singularities occur on a proper or an improper cone. Putting q = S, r=l, it follows that the additional reduction of class produced by a triple generator on the nodal cone is 4 ; and it can be shown, by an independent investigation, that the DEGRADATION OF NODAL CONE INTO PLANES 139 additional reduction of class produced by a triple generator when the nodal cone is a quartic cone is the same, whether the cone is (i) a proper one, (ii) a nodal cubic cone and a plane through the nodal generator, (iii) two planes and a quadric cone passing through their line of intersection. Multiple Points in which the Cone co7isists of Planes intersecting in the same Straight Line. 196. We shall now discuss multiple points in which the cone degrades into p planes intersecting in the same straight line, and shall commence with the following theorem. When a multiple point of order p consists of p distinct planes intersecting in a point, the reduction of class is ^p{p-l)(2p-l), and the point constituents of the singularity are G = ip(p-l)(p-2), B = ^p(p-1). But when the planes intersect in the same straight line, the reduc- tion of class is (p +l)(p — ly, and the point constituents of the singularity are c=i(p-i)Hi>-2), B=(p-iy. By means of the theorem of § 24, it can be shown in both cases that the reduction of class does not arise from the union of any additional conic nodes or binodes with the multiple point ; hence the reduction is caused by the conversion of conic nodes into binodes. In the first case, when the planes intersect in the same point, the number of their lines of intersection is ^p (p — 1); hence 2G + SB=p(p-iy + ^p{p-l), also 0+ B=lp{p-lf, whence G =^p{p—\){p — 2), B = ^p(p—1) (1). To prove the second case, we may employ a surface of degree p + l, which is avp + Up+i = (2), where m^+i = /3^+%o + /S^^^i + . . . Wp^^ . The first polars of (7 and JD are av'p + u'p+i = 0, av"p + m"^+i = (3), 140 SINGULARITIES OF SURFACES Multiplying the first of (3) by 7 and the second by B and adding, we obtain pavp + ^Pw, + 2/SP-'w, + ...{p + l) Wp+, = (4). Eliminating a between (2) and (4), we obtain p/3P+^Wo + ip-l)l3Pw,+ ...-iVp+, = (0). Eliminating a between (3), we obtain (y8^< + y8^-W+ ...) V' = (/SP<' + /3*-^w;'+ ...)<.. .(6). Equations (5) and (6) represent two cones of degrees p + 1 and 2p—l, which possess (p + 1) (2p — 1) common generators; and this number is equal to the number of ordinary points of intersection of (2) and the first polars of two arbitrary points. Hence m = (p + l) {2p -l) = (p + l)p^-2G- HB, accordingly 2C + dB = (p+ l){p-iy (7), also G+ B = lp(j)-\f, whence C=\{p-lf {p -2), B = {p-\f (8). 197. When s tangent planes coincide, the reduction of class is (p + l){{p-lf + s-l]. In this case Vp = 8^Vp^s', hence V = ^'v'p-s, V = ^'~' {^f^"p-s + svp-s) ; accordingly (6) contains S*~^ as a factor which must be rejected, and the resulting cone is of degree 2p - s. Whence m = (p + l){2p - s) = (p + l)p^ -2G - SB, giving 2C + SB = (p + l){{p-iy + s-l} (9), from which it follows that each successive coincident plane pro- duces an additional reduction of class equal to p + 1. When all the planes coincide, p = s, and the reduction becomes p (p^ ~ !)• 198. We shall now explain a method for determining the number of constituent conic nodes and binodes, when some of the planes coincide, which depends upon the theorem of § 24. Two cones of degree n which have a common vertex possess n^ common generators ; and a pair of such cones may be regarded as an improper cone of degree 2n which has 01^ nodal generators. If the two cones have an additional common generator they must coincide ; hence a cone of degree n twice repeated may be regarded DEGRADATION OF NODAL CONE INTO PLANES 141 as an improper cone of degree 2n which has n^ + 1 nodal generators. From this it follows that a pair of coincident planes may be regarded as a hinodal quadric cone, the positions of whose nodal generators are indeterminate. Similarly three coincident planes may be regarded as an improper cubic cone having 2 + 2 + 2 = 6 common generators ; and generally if ts be the number of nodal generators when there are s coincident planes, the solution of which is ts = s{s — l). Accordingly s coincident planes may be regarded as a cone of degree s which has s{s — \) nodal generators, the positions of which are indeterminate. 199. When s coincident planes coincide, the constituents of the singularity are G=^\{p-\f{p-2)-{p+l){s-^l B = {p-\Y + {p + l){s-^); when p + \^s{s-l), and C=^{p-\f{p-^)^{s-\){^s-p-l), B = (p-iy-{s- 1) (2s -JO - 1) ; ivhen p+1 ^s(s-i). The equation of the surface is aP-PB%-s + a«-^-%p+i + ...Un = (10). In the first case, each of the p + 1 lines of intersection of the cone Up+i with the plane 8 may be regarded as nodal generators of the cone 8* ; hence the number of additional nodes is p + 1, and G + B = ^p(p-iy+p + l (11). Combining this with (9) we obtain the first result. But in the second case there are only s{s — 1) additional nodes, whence G+B = lp{p-iy + s(s-l) (12), which by virtue of (9) gives the second result. When p = s, p + l2, the proper formulae are the first ones, and we obtain G=i(p-inp-2)-(p+i){p-4.)l B = (p-iy + (p + i){p-s) J ^'"'^' where p > 2. 142 SINGULARITIES OF SURFACES 200. I shall now explain another method of finding the reduction of class produced by a multiple point. Let A he a multiple point of order p, the tangent cone at which is anautotomic, then the class of the cone is p(p -1); hence if I) be any point of space, p{p — 1) tangent planes can be drawn to the cone through DA, and therefore DA is a multiple generator of the tangent cone from D of order p{p — '[). Now the class m of the tangent cone from D is the same as that of the surface ; hence by § 167, m — 2p(p — 1) tangent planes can be drawn to the surface through the line DA. This number is obviously equal to the class fi of the tangent cone from A to the surface ; hence m — 2p{p — l) = fi (1), which reduces the problem to the determination of the class of the tangent cone from A. Let the surface on which the multiple point exists be of degree p + 2; then since none of the generators of the tangent cone from A can be double or stationary tangents to the surface, it follows that the tangent cone is anautotomic ; and since the equation of the surface is a^Lip + 2a^tp+l + it^+a == 0, that of the tangent cone from A is ^ p+l — UpUp^2 (■^)) whence its degree is 2p + 2, and its class fx, = {2p + 2) (2p +1). Substituting in (1), we obtain m-2p{p-l) = (2p + 2)(2p + l) (3), giving m = {p + 2)(p + iy—p{p — l)% and the last term is the reduction of class produced by the multiple point. Equation (2) shows that the lines of closest contact are generators ; hence a singular generator of Up which is not a line of closest contact will not affect the value of /a, but the left-hand side (1) will be altered. Let the nodal cone at A possess 8 nodal and K cuspidal generators, then the section of this cone by the plane BCD will be a curve of degree p having S nodes and k cusps, and the number of tangent lines which can be drawn through D to the section is p(p — l) — 28 - 3«, and this is consequently the number of distinct tangent planes which can be drawn through DA to the SINGULAR GENERATORS OF NODAL CONE 143 nodal cone. But every plane through DA and a nodal generator is equivalent to two coincident tangent planes, and every such plane through a cuspidal generator is equivalent to three coincident tangent planes. Hence DA is a singular generator of the tangent cone from D of order p{p — l) having S pairs of coincident tangent planes, corresponding to each nodal generator, and k planes con- sisting of three coincident tangent planes which correspond to each cuspidal generator. Putting r = 2 and s = 3 in (9 A) of § 167, it follows that the number of tangent planes which can be drawn through DA to the tangent cone from D is m- 2p(p -1) -\- {r - 1) S + (s - 1) fc = m-2p(p -l) + 8 + 2fc; and since this is equal to the number of tangent planes which can be drawn through DA to the surface, (3) must be replaced by m- 2p(p -1) + B + 2K = (2p + 2){2p + 1), giving m = (p + 2) (p + ly — p (p — ly — B — 2k, which furnishes another proof of the theorem of §194, 201. In I 169 we have called attention to the distinction which exists between multiple points and singula?- points on plane curves ; and we shall now prove that : If the nodal cone at a multiple point of order p possesses a singular generator of order 2, whose constituents are 8 nodal and k cuspidal generators, which move up to coincidence along a continuous curve, the total reduction of class is p{p-\f + 2B + ^K-l and the point constituents are G = \p{p -If -2B-U + 1, 5 = 284-3/c-l. In this case, the number of distinct tangent planes which can be drawn through DA to the nodal cone oXA i?, p{p — l) — 2B — 3/c as before ; but the number of coincident tangent planes is 2S + 3/c. Hence the number of tangent planes which can be drawn through DA to the surface is m — 2p(p — 1) + 2S + 3/c— 1. Substituting this quantity for left-hand side of (3) we obtain m = {p + 2){p + lf-p{p-rf-2h-^K + \. Also since the value of G-\-B is given by (20) of § 194, the theorem at once follows. Let us write 5 = 2S + 3« — 1, then the following special cases 144 SINGULARITIES OF SURFACES may be noted when the nodal cone has the following singular generators : (i) Tacnodal generator. Here 5 = 2, /c = ; whence 5 = 3. (ii) Rhamphoid cuspidal generator. Here S = 1, k = 1; whence 5 = 4. (iii) Oscnodal generator. Here S = 3, k = 0; whence s = 6. (iv) Tacnode cuspidal generator. Here S = 2, k=1\ whence 5=6. 202. The preceding theorem requires modification when the singular generator is a line of closest contact ; and we shall show that: If AB is a singular generator of order 2 on the nodal cone, which produces an additional reduction of class equal to s ; then when AB is a line of closest contact, the additional reduction is 5 + 1, and the point constituents of the singularity are G = ^p(p-iy-s + 2, 5 = 5-1. The equation of the surface must be of the form ^n-p (^p-2 ^^ ^ ^p-3 y^ + ^ ^ . ) _,_ f^n-p-i (^p+l W0 + /3PW2+...) + a^-P-^Up+^ + ... Un = 0, where the w's are arbitrary binary quantics of (7, S), but the vs are connected in a manner which depends on the character of the singular generator AB. The latter has {p + l)-tactic contact with the surface at A, and 2?- tactic contact with the first polar at A ; but if Wo=0, so that AB becomes a line of closest contact, then AB has (p + 2)-tactic contact with the surface at A and {p + V)- tactic contact with the first polar at A. This shows that the surface and the first polars of any two points intersect in an additional point at A ; hence the total reduction of class is 2G + W=p{p-iy-{-s+\. Also by § 24, this additional reduction of class is produced by an additional double point which moves up to coincidence with A ; . accordingly c^B=ip{p-ir+\, which gives the required result. The reader will be assisted in understanding the process which takes place, by considering the case of an ordinary conic node. When the nodal cone has a nodal generator (that is becomes two SINGULAR LINES OF CLOSEST CONTACT 145 planes) the conic node is converted into a binode ; but when this generator becomes a line of closest contact, the binode is reconverted into a conic node, and an additional conic node added, so that the singularity becomes the special binode whose axis has quadri tactic contact with the surface. 203. The following theorem is an extension of this result. If the nodal cone at a multiple point of order p possesses B nodal generators, each of which is a line of closest contact ; then when all coincide, the constituents of the singularity are G=^p(p-iy-B + 2, 5 = 28-2. To prove this theorem it will be sufficient to consider the case of a tacnodal generator on a quartic cone. When two generators AB, AB' are lines of closest contact the effect is to add two conic nodes to the constituents of the singularity; so that in the case of a quartic node, the total reduction of class is 2(7+ 35 = 36 + 2 + 2 = 40 ; and we shall now show that when AB and AB' coincide, the effect is to produce a further reduction of class equal to 2. Consider the surface a (/3V + 2^ViV2 + 2^4) + k^% + /3%2 + . . . W5 = (4), where Vi = 7 + S and A; is a constant. Equation (17) of §194 now becomes {^W + 2y8viW2 + 2^4) (A;/3* + /3 V + • • • O = 2 {k^% + /3^W2 + . . . w,) {^% + /3(v, + v,v^) + v:] (5), or k^W + ^'^1 (2A;yi< + 2w^ - kv^w^) + . . . = (6). Now write down the equation corresponding to (18) of § 194 and subtract, and it will be found that we shall obtain an equation of the form ySVXii+yS%A + /S'n5+...n8 = o (7), from which it is easily shown* that (6) and (7) intersect in 54 ordinary generators and in 10 coincident generators along AB. * Eegarding (6) and (7) as plane curves, we have to find the number of coincident points in which they intersect at B ; and we may replace them by two equations of the form /3V + /37V2 + 1^4 = 0, ^^y^Wi + PjWs + W5 = 0. Eliminating ^37, we obtain a binary decimic of (7, 8) which shows that the two curves intersect in 10 ordinary points, and therefore the number of points absorbed at 5 is 20-10 = 10. B. 10 146 SINGULARITIES OF SURFACES But since AB is four times repeated amongst the lines of closest contact of (4), the total number of ordinary lines of intersection of (6) and (7) is 64 -10 -(20 -4) = 38, whence m = 38 = 80 - 26' - SB, giving 2C+35 = 42 = 36 + 6. From this it follows that the coincidence of the two generators AB, AB' produces a further reduction of class equal to 2 ; and by taking a third generator AB", which is a nodal generator on the cone and is also a line of closest contact, it can be shown that an additional reduction of class 2 + 2 = 4 is produced. Generalizing it follows that when there are 8 coincident nodal generators, all of which before coincidence are lines of closest contact, 2G+SB= p{p-iy + ^8-2, 0+ B = ^p{p-iy + 8, which proves the theorem. Cubic Nodes. 204. There are six primary species of cubic nodes. I. In the first species the nodal cone is an irreducible cubic cone. Of these there are three subsidiary species, which occur when the cone is (i) anautotomic, (7=6, B = 0; (ii) nodal, 0=5, B = l; (iii) cuspidal, G = 4, 5 = 2. II. In this species the cone consists of a quadric cone and a plane ; and there are two subsidiary species according as the plane (i) intersects the cone in two distinct generators, (7=4, 5=2; or (ii) touches the cone, G = S, B = S. In the latter case the cone is a reducible or improper cubic cone having a tacnodal generator, and the values of the constituents follow from § 201. III. Three planes intersecting in a point, G = d, B=S. IV. Three planes intersecting in the same straight line, (7=2,5=4. V. One distinct and two coincident planes, (7=4, B = 4t. VI. Three coincident planes, (7=6, 5 = 4. All these results follow from the preceding theorems. With regard to V, it follows from § 199 that p = S, 5 = 2, so that the QUARTIC NODES 147 second formulae are the proper ones ; but in the case of VI, p = s = S and the first formulae must be used. 205. It will be noticed that I (iii) and II (i) have the same point constituents, and a similar remark applies to II (ii) and III. Exactly the same thing occurs in the theory of plane curves, for the point constituents of a triple point of the second kind and of a tacnode cusp are both equal to 8 = 2, k = 1; but the line constituents are different, for in the former case they are T = 0, 1 = 0, and in the latter t = 2, t = l. And since surfaces possess plane as well as point singularities, it is practically certain that the plane constituents in the above respective cases are different; although the theory has not yet been worked out. Quartic Nodes. 206. The theory of quartic nodes is coextensive with that of quartic curves, since a plane section of the nodal cone may be any quartic curve proper or improper. The theorems of § 201 give the reduction of class when the nodal cone has a tacnodal, a rhamphoid cuspidal, an oscnodal and a tacnode cuspidal generator which is not a line of closest contact ; the theorem of § 202 solves it when the generator is a line of closest contact ; whilst that of § 203 solves it when the cone m^+i touches the nodal cone along a tacnodal generator, or osculates it along an oscnodal one. Triple generators are discussed in § 195, and the various cases in which the nodal cone degrades into four planes are dealt with in §§ 196 — 9. I shall therefore only discuss two additional cases for the purpose of illustrating the method employed. 207. When the nodal cone consists of a quadric cone and two coincident planes, the point constituents of the quartic node are C=l% B = 8. We shall employ the sextic surface + B'W,+ ... We = ...(1), where the suffixed letters denote binary quantics of {^, y). Write (1) in the binary form (a, b, c, d, e^B, 1)* = 0, 10—2 148 SINGULARITIES OF SURFACES then the equation of the tangent cone from D is I' - 27 J2 = 0, where I = ae — 46c? + 3c^ / = ace + 2bcd — ad? — Ife — c^, and the values of a, b, c, d, e are a = a^ + aFi+ W^, c = ^a2/37 + aFg + W„ d = oiV,+ W„ e = aF,+ Fe. Writing down the discriminantal equation of the tangent cone from D, it will be found that the highest power of a is a", and the term involving it is 27c^e (3ac — 26^); and that its coefficient (re- jecting constant factors) is fi^'fV^{^^ — ^v^); also the coefficient of a" does not involve /8 or 7 as a factor. From this it follows that if fjb be the class of the cone, the number of tangent planes which can be drawn through DA to the surface is /i — 22 = w — 26, since yu, = m — 4. The tangent cone /rom A has five nodal generators, which are the lines of intersection of the plane h and the cone Fg ; also AD is another nodal generator of the tangent cone ; hence its class is 90 — 10 — 2 = 78. The number of tangent planes which can be drawn to this cone, and therefore to the surface, through AD is accordingly 78 — 4 = 74; and we thus obtain m — 26 = 74, giving m=100. Let X be the reduction of class produced by each of the lines AB and AD, then m = 100 = 150 - 36 - 2a; - 2 - 2 - 2, giving a; = 4. Since the surface possesses an isolated conic node at D, it follows that if this were absent we should have m = 102, whence 2(7 + 35 = 48. Also from § 198 the union of the two planes produces two additional conic nodes, whence C + 5 = 20, from which we obtain C =12, B=8. SINGULAR LINES AND CURVES 149 208. When the nodal cone at a quartic node consists of a quadric cone twice repeated, the constituents of the singularity are a=15, 5 = 8. Consider the surface oiu^ + Wg = 0, then, proceeding according to the first method, the two cones whose common generators determine the class are the sextic cones U^U^ = '^2^5, which possess 36 common generators; but, since 10 of these common generators are the 10 lines of closest contact, which are the common generators of the cones u^ and u^, the total number of ordinary generators is 26. Hence 7n = 26 = 80 - 2C - 2>B, whence 2(7+35=54. Now, from § 198, it appears that a quadric cone twice repeated may be regarded as a quinquenodal quartic cone whose nodal generators lie on the quadric cone u^, but are otherwise inde- terminate ; also these five nodal generators may be regarded as coinciding with five of the lines of closest contact. Hence C+5=18 + 5 = 23, giving 0=15, 5 = 8. The foregoing result is capable of extension to multiple points whose nodal cones are of the form u^Uq. Singular Lines and Curves*. 209. A surface may possess any line or curve lying in it, such that an arbitrary plane section through any point P on the line or curve has a singular point at P of any species which an algebraic plane curve can possess. Moreover all singular lines and curves possess singular points, analogous to pinch points, at which the singularity changes its character. Thus a cuspidal line possesses certain points at which the cusp changes into a tacnode, and a triple line of the first kind points at which the triple point changes into one of the second kind and so on. We also saw in § 41 that * Basset, Quart. Jour. vol. xxxix. p. 334. 150 SINGULARITIES OF SURFACES nodal lines of the third kind possess cubic nodes but no pinch points ; and in like manner it will be found that singular lines and curves possess multiple points lying in them of a higher order of singularity than that of the line or curve. The tangent planes at any point on a multiple line are in general tarsal tangent planes ; it is however possible for any tangent plane to be fixed in space, and such lines usually possess distinct features of their own. There are consequently two species of cuspidal, tacnodal &c. lines, in the first of which the tangent plane is torsal and in the second it is fixed in space. Cuspidal Lines. 210. The general equation of a surface having a cuspidal line of the first species is (Lay + M^Bf (a, |8)-^ + (P, Q, R, ^$7, 8)^ = (1), where P, Q, R, S are quaternary quantics of (a, ^, 7, S) of degree n—S. This equation when written out at full length is (Lay + M^Sy {poo^-' +piOL''-'j3 +... i?n-4/3"~') + ol'^'^ Cuspidal lines possess two kinds of singular points which occur (i) when the cusp changes into a tacnode, (ii) when there are cubic nodes on the line. Let the plane a = \^ cut AB in B', then the equation of the section is /3'»-2 {L\y + MBf (poX""-' + piX""-' +... pn-i) + jS""-' (V'-% + X""-' F3 + . . . W3) + . . . = . . .(3), which, for brevity, we shall write in the form ^/3'*-2ni2 + 5/3'*-3+... = (4). 211. The cuspidal line possesses n tacnodal points, and n — 4 cubic nodes, at which there is a cuspidal cubic cone. The condition for a tacnodal point is that Hi should be a factor of B, which requires that the eliminant of Hi and B should vanish. This furnishes an equation of the nih. degree in \. The points where the cuspidal line cuts the planes (a, /3)'*~* = are cubic nodes on the line, and there are 7i — 4 of them. If A be one of these points p^ = 0, and the coefficient of a^~^ equated to zero gives L%^y' + v, = 0, CUSPIDAL LINES 151 which shows that AB is a cuspidal generator of the cone. A cuspidal line on a quartic surface has 4 tacnodal points, but no cubic nodes ; hence on such a surface the line appears in an in- complete form. 212. A cuspidal line of the second species possesses n — S tacnodal points and n — 2 cubic nodes. If 7 be the fixed tangent plane, the equation of the surface must be of the form y^{a,^r-^ + {P,Q,R,Slry,Sr = (5). Proceeding as in § 210, the first term of the equation corre- sponding to (3) must be of the form ^n-2^2 [p^Xn-2 +^^X"-3 + . . . _p^_^j^ and the condition for a cubic node is that this should vanish, which furnishes an equation of degree ?i — 2 in X. The condition for a tacnodal point is that the coefficient of B^ in the expression should vanish, which furnishes an equation of degree n — S in X. A quartic surface having a cuspidal line of this character possesses both species of singular points. 213. The discussion of other species of singular lines is very similar, and I shall therefore merely give the results, referring the reader to my paper on Singular Lines and Curves on Surfaces. Tacnodal Lines. The equation of a surface having a tacnodal line of the first species is (Lay + M^Bf (a, /S^-' + 2 (Lay + if/35) (F, G, ... ][a, yS)'*-* + {P,Q,R,8,Tly,Sy = 0...(6), where F, G, ... are binary quantics of (7, 8) ; and P, Q, R, S, T are quaternary quantics of all the coordinates. The singular points consist of (i) 2ri — 4 points where the tacnode changes into a rhamphoid cusp; (ii) w — 4 points which are cubic nodes, the nodal cone at which consists of a quadric cone and a plane touching the latter along the tacnodal line. The equation of a tacnodal line of the second kind is 7' («, /3)"~' + 2y{F,G... Ja, /3y-'' + iP,Q,R,S,T^y,Sy = 0...{7), 152 SINGULARITIES OF SURFACES and it possesses 2n — 6 rhamphoid cuspidal points and n — 2 cubic nodes. Rhamphoid Cuspidal Lines. The equation of the surface when the line is of the first kind is (Lay + M^B + pry^ + qyS + rB'f (a, ^f-^ + {Lay + M^S)(F, 0, ...^u,^r-^ + (P,Q,R ...^y, Sy = 0...{8), where F, G, ... are binary cubics of (7, S), and P, Q, ... are quaternary quantics of all the coordinates. These lines possess n — 4i cubic nodes, and n points where the rhamphoid cusp changes into an oscnode. When the line is of the second kind, its equation is (ay + 297' + ^7^ + rBy (a, ^y-"- + y^ (a>'-\ + a^-'/3w^ + ...) + 7 (a"-%3 + OL^'-'lSws +...) + a^-'w, + OL^-'{^W,+ We)+... Wn = 0...(9), and the line possesses w — 4 cubic nodes and n — S oscnodal points. 214. The highest singular line of the second order and first species which a quartic surface can possess is a tacnodal line ; but when the line is of the second species, such a surface may possess a rhamphoid cuspidal and an oscnodal line. The equa- tions of the surface in the two respective cases may be reduced to the forms {ay+B'y + y(ayVi + ^yw^ + Ws) = (10), and (ay + BJ + y' {Pa+ Ql3 + Ry + SB)= (11). The section of (11) by the plane a = X/3 is (X^y + B^y + y' [{P\ +Q)^ + Ry + 8B] = 0, and the condition that B' should be a tacnode cusp is that \= — QjP. An oscnodal line of the second kind on a quartic has therefore one tacnode cuspidal point on it. It is not possible for a quartic to have a tacnode cuspidal line, since the conditions are that PX + Q = for all values of \, which require that P = Q = 0, in which case (11) becomes a cone. Triple Lines. 215. There are ten primary species of triple lines. I. Three distinct tangent planes ; all of which are torsal. II. Three distinct tangent planes ; one fixed and two torsal. TRIPLE LINES 153 III. Three distinct tangent planes ; two fixed and one torsal. IV. Three distinct tangent planes ; all three fixed. V. Two coincident fixed tangent planes ; one distinct torsal plane. VI. Two coincident fixed tangent planes ; one distinct fixed plane. VII. Three coincident fixed tangent planes. VIII. Two coincident torsal tangent planes ; one distinct torsal plane. IX. Two coincident torsal tangent planes ; one distinct fixed plane. X. Three coincident torsal tangent planes. Triple lines possess a variety of species of singular points, which we shall proceed to consider. Thus when the line is of the first kind, points exist at which a pair of tangent planes coincide, so that the section of the surface through the point has a triple point of the second species thereat ; also in certain cases points exist which are quartic nodes on the triple line. I. A triple line of the first kind on a surface of the nth degree has 4n — 12 points at which two of the tangent planes coincide. The equation of the surface is of the form (P,Q,R,Sly,Sr=0 (12), where P, Q, R, S are quaternary quantics of degree n — S. Equation (12), when written out at full length, becomes d^-% + a"-* (/3w3 + w,) + ... /S""-' Ws + /3''-* W, + ...Wn = 0.. .(13), and the equation of the section by the plane a = X./3 is ^n-3 (x'^-3^3 + \^-*Ws + ...Ws) + ^'^-^ (\"-%4 + X*^-' F4 + . . . F4) + . . . = 0. . .(14), or ^/3^-^ + 5/3'*-*+... = (15). The points at which a pair of tangent planes coincide will be called pinch points, and the condition for their existence is that the discriminant of A should vanish; and since ^ is a binary cubic of (7, 8) whose coefficients are polynomials of X of degree w — 3, the discriminant is of degree 4n ~ 12 in X. Every tangent 'plane touches the surface at n — 3 distinct points. 154 SINGULARITIES OF SURFACES The condition that 7 should be a tangent plane at the point B' where the plane a = \/3 cuts AB, is that the coefficient of B^ in A should vanish. This furnishes an equation of degree n — 3 in \, which shows that there are w — 3 of such points. The theory of coincident pinch points will be considered in § 216, but in the meantime I shall enunciate the theorems concerning them. II. When 2w — 6 pinch points coincide in pairs, one of the tangent planes is fixed in space, and the line becomes one of the second species. III. When all the pinch points coincide in pairs, two tangent planes are fixed in space. IV. When all the tangent planes are fixed in space, the triple line possesses n — S quartic nodes. In this case A = yBv, {V'-^ + ...Wo) = yBv.Ao, and a quartic node will occur whenever A^ vanishes. It will hereafter be shown that the pinch points coincide in quartettes at each quartic node. The equation of the surface may be written in the form jBv, (a, /3)-^ + (P, Q, R, 8, T^r^, By = 0, and the points where quartic nodes occur are given by the equa- tion (a, ^Y~^ — ; hence if J. is a quartic node, the nodal cone is of the form /37SW1 + V4 = 0, which is the equation of a quartic cone having a triple generator of the first kind. V. When all the pinch points coincide in quartettes, two coincident tangent planes are fixed in space, and the third one is torsal ; also the line possesses n — ^ singular points, at which the triple point of the second kind changes to one possessing a pair of tacnodal branches and one distinct ordinary branch. The value of A in equation (15) is of the form A=y'B„ and the condition for a pinch point is that the coefficient of B in Bi should vanish, which shows that there are n — 3 apparent pinch points. TRIPLE LINES 155 Equation (15) now becomes B,^''-Y + B^''-' + . . . = 0, and if the coefficient of S in 5 vanishes, 7 will be a factor of B, and the point consists of a pair of tacnodal branches and one ordinary branch through it. Since B is of degree n — 4 in X,, there are n — 4 of such points, and, like pinch points, they affect the class of the surface. VI. When two coincident tangent planes are fixed in space, and the distinct plane is also fixed, the triple line possesses n—S quartic nodes, and n — ^ of the points considered in V. For the equation of the section by the plane a. = \j3 is {V'-% + ... Fo) yS^-^Y^S + B^^-^ + . . . = 0. VII. When three coincident tangent planes are fixed in space, the line possesses n — 3 quartic nodes, and w — 4 points consisting of a pair of tacnodal branches and a coincident ordinary branch. The equation of the section is AS''^'^^ + -S/S'*-' + . . . = 0. The first kind of points occur when Aq = {), and the second when 7 is a factor of B. The constituents of the latter point (on a plane curve) are three nodes and one cusp ; and both kinds of points affect the class of the surface. 216. The theory of coincident pinch points is best investigated by the following method. Let A be the discriminant of (12), so that A = P'B' - 6PQRS + 4^PR' + ^Q'S - SQ'R' (16), then A = is a surface of degree 4/i — 12, and we shall first show that the pinch points occur where AB intersects the surface A = 0. Let P = PoOi^-' + P,a^-' + . . . P«_3, with similar expressions for Q, R, S, where Pns = {^, y, 8)"~^ ; then if J. be a pinch point and y'^S = the equation of the tangent planes thereat, it follows from (12) that Po — Ro = So = 0. The term 4Q^^ in (16) contains the highest power of a which is the (4/1 — 13)th, and shows that the surface A = passes through A. In the next plane consider the line II in which 7 = is the 156 SINGULARITIES OF SURFACES fixed tangent plane. The values of P, Q, and R remain unaltered, but S=ty + TS, where %T= (a, /3, 7, By-\ Let t = a-oa''-' + o-ia"-s + . . . o-„_4, where cr^, tn = {^, 7, S)**; then the two terms in (16) which con- tain the highest powers of a are putting 7 = S = 0, this reduces to A?yQV*~^*, where A; is a constant, which shows that the line AB touches the surface A = at A, and therefore two pinch points coincide. If however we had supposed that the tangent planes at A were yS^, we should find that A = intersects but does not touch AB at J., so that A is an ordinary pinch point. Accordingly the discriminantal surface cuts AB in 2n— Q points and touches it at ti — 3 points, which shows that there are 2^1 — 6 distinct pinch points and 2n — 6 which coincide in pairs. In the same way any other case may be treated. We shall now inquire what becomes of the pinch points in the case of line IV. Let if be a binary quantic of (a, /3) of degree n — S; also let P, P', &c. be quaternary quantics of degree n — 4. Then the equation of the surface may be written in the form rf (Py + Q8) + SyS (Mf+ Fy + Q'h) + 875^ {Mg + R'y + 8'B) + h' (Ry + SS) = 0, where 7S (/y + gS) = are the three fixed tangent planes. Writing down the equation of the discriminantal surface, and then putting ry = 8 = 0, it will be found to reduce to — Sf^g^M^ = 0, which shows that the surface A has quadritactic contact with the line AB 8bt the quartic nodes. This shows that the pinch points coincide in quartettes at the quartic nodes. 217. VIII. The remaining three species present many features in common with cuspidal lines of the first kind ; and the equation of a surface having a line of the eighth species is (Lay + M^hy (F, G,... $«, /3)»-« +{p,Q,R,s/riy,hy=Q ...(17), where F, G, ... are linear functions of (7, h), and P, Q, ... are quaternary quantics of (a, /3, 7, h). Equation (17) when written out at full length is of the form TRIPLE LINES 157 {LoL'y + M^hf (Fa''-' + Ga^'-'IS + ... K^^-') + a'^-%4 + a''-' (/3 F4 + Fg) + . . . + /S'^-%4 + ...Wn = 0.. .(18). Equation (18) shows that no surface of a lower degree than a sextic can possess a line of this species ; for if n = 5, (18) reduces to the form (Lay + M^Byvi + aVi + ^Wi + w, = (19), in which the distinct tangent plane is fixed in space, and the line therefore belongs to species IX. The section of (18) by the plane a = X/3 is (L\y + MSy (FV'-' + GX""-' +...) /3''-' + (V'-% + V'-'V,+ ...)^^-*+...=0, which we shall write Avi'f3''-' + Bl3^-'+... = (20). (a) The first kind of singular point occurs when all the tangent planes coincide, and there are n — 4< of them. The condition for these points is that A = kv-^, where ^ is a constant, which furnishes an equation of degree r? — 4 in \. (h) There are n points at which the triple point of the second kind changes into one consisting of one pair of tacnodal branches and an ordinary branch passing through it. The condition for these points is that v^ should be a factor of B, which furnishes an equation of degree n in \. IX. When the distinct tangent plane is fixed in space, the equation of the surface is {Lay + MjSBf (py + qB) (a, /3)"-^ + (P,Q, R,S,T'^y, By = 0... (21), and the section by the plane a = \/3 is (L\y + MBf (py + qB) (FX""-' +...) ^^-^ + (V'-% + X''-'V, + . . .) /S**-" + . . . = 0, or Av^'w,/3''-^ + B^^-^+...=0 (22). (a) There are n— 5 quartic nodes which are the intersections of the triple line and the planes (a, I3Y~' = 0. A quintic surface cannot possess these quartic nodes, and therefore the singularity occurs on such a surface in an incom- 158 SINaULARITIES OF SURFACES plete form. When n> 5 the equation of the nodal cone is of the form 7^ (py + qB)l3 + Vi = 0. (b) There is one pinch point, which occurs when Lq\ = Mp. (c) There are n points luhere the triple point of the second kind changes into one consisting of a pair of tacnodal branches and one distinct ordinary branch. X. The equation of the surface is {LoLr^ + M^ZYia, ^r-' + {P, Q, R, S, T][7, Sy = 0...(23), and the section by the plane a = X/3 is {LXy + M8f (FX''-' +...) 13^-^- + {\''-% + . . . ) y8'*-^ + . . . = 0, or A iS^'-'v,^ + B^-^ + . . . = 0. (a) There are n — Q quartic nodes, and the equation of the nodal cone is of the form rf^ + V4 = 0. A sextic is the surface of lowest degree which can possess this line, and since there are no quartic nodes the singularity occurs in an incomplete form. (b) There are n points at which the triple point of the third kind changes into one consisting of a pair of tacnodal branches and one coincident ordinary branch. Both these singular points affect the class of the surface. Nodal Curves. 218. The equation of a surface of the nih. degree which has a plane nodal curve of degree s is a^ Vn-2 + 2ansUn-s-i + n/ w,i_2s = (24), where F is a quaternary quantic of (a, y8, 7, B), and D,, u are ternary quantics of (/3, 7, S) of the degrees indicated by the suffixes. We shall usually omit the suffix s in 12. 219. The nodal curve possesses 2n (71 — s — 1) pinch points, which are the points of intersection of the curve and surface ''' ji— s— 1 ^^ ' n—-i^n—is (.•""/• Let B be one of the points of intersection of (25) and the NODAL CURVES 159 nodal curve, and let u^-^ be the portion of Vn-2 which is inde- pendent of a ; then (25) may be replaced by '^ n—S-l ^^ ^("11—2^11—28 V^"/* Since S = is any arbitrary section of the surface through B, we have to show that the section has a cusp at B ; also since (26) has to pass through B which is a point on 11, it follows that when S = Un-2 = f^''-^ + PiyS"-' 7 + • • • i?«-27"~'] Un-s-, = MyS--^-^ + ^,yQ'^-«-^ 7 + . . . ( ^27). Un-2B = q'/S""-^ + n^""-''-' 7 + • • • I" From these results it follows that the highest power of /3 on the section of (29) by S is the (n — 2)th and that its coefficient is (pa. + qkf{f, which shows that 5 is a pinch point. 220. The plane a intersects the surface in the nodal curve twice repeated, and in the curve a = 0, Un-2s = 0, which is called the residual curve ; and the latter curve intersects the nodal curve in s{n—2,s) points. When the curve is of a higher order of singularity, these points are as a general rule multiple points on the singular curve, but when the latter is nodal the plane a is a tangent plane at these points ; in other words, a is one of the two nodal tangent planes. To prove this, let B be one of the points in question ; then where Un = (a, 7, S)", and v^ = w„ = 0„ = (7, 8)^ from which it follows that the coefficient of yS'*-^ in (24) is a (auo + 2vofli). Cuspidal Curves. 221. When the singular curve is cuspidal, every point on it must be a pinch point, which requires that equation (26) should contain H as a factor. Accordingly U n—s—i '^n—2^n—2s ^^ i^ ' i' • yZo)y 160 SINGULARITIES OF SURFACES where 0,' = (y8, 7, 8)2»i-3s-2, xhe right-hand side of (28) vanishes at every point on the cuspidal curve, hence the left-hand side of (28) must do so also. Now Un-2s vanishes at the points where the residual curve intersects the cuspidal curve, hence Un-s-i must also vanish at these points ; accordingly where o- and w are undetermined ternary quantics of (/S, 7, 8). Substituting from (29) in (28), we obtain and since O has to be a factor of the left-hand side, we must have Un-2 = Un-2s(^\-i + 0.n-s-2 (30), where +,) + a''-P-'Up+, + ...Un = (11). By means of the methods explained in Chapter IV, it can be shown that the section of (11) by the plane Vi = Ti is a curve having a rhamphoid cusp at A and p - 2 ordinary branches through it. To find the reduction of class, it will be sufficient to consider a quartic surface, since the method employed is applic- able to any surface. Putting n = 4<, p = 2, Wi = B, Tj = J07 + qB, the surface (11) becomes (aS + WjS') [a (py + qS) + W^'] + a (^w, + w,) + ^'W^ + ^'W, + I3W,+ 1^4 = 0.. .(12), and the first polar of D, which may be any arbitrary point, is a {a (py + qB) + W^'] + qot (aB + W^') + a {/3w^' + <') + /S='F/'+... W:' = 0...iU). The sections of (12) and (13) by the plane B=py -{- qB may be written in the form {aB+W^J + oiB^A, + BA, = (14), and a{aB+W/3^) + a8B, + B, = (15), where A^ = Bn = (/S, By. Eliminating a between (14) and (15), it will be found that the eliminant is a binary septimic of (/3, B), which shows that (14) and (15) have quinquetactic contact with one another at A. Accordingly the surface (12) and its first polar with respect to any arbitrary point intersect one another in five coincident points at A ; hence the reduction of class is 5, which shows that the singularity at A is formed by the union of a conic and a binode. The case of a cubic surface is peculiar. Let us consider the equation 0LU2+Us = (16), which represents a cubic surface having a conic node at A. The transformed surface, when written at full length, becomes a (^'Wo + I3'w, + ySwo + Ws) + n (/3X + ^v, + v^) = . . .(17). In order that (17) should have a conic node at B, it is necessary that Wo = Va = Vi = 0, in which case the equation of the nodal cone is awi + ^2 = 0, BIRATIONAL TRANSFORMATION 169 and, if 5 is a binode, V2 = w^Ti and (16) becomes awjTi + ^^Wi + ^W2 + Ws = (18), and the singularity at A is Salmon's binode Br,, which, as pointed out in § 80, is a singularity of a different character to the singular point 0=1,5=1 on a surface of higher degree than the third. 232, The preceding results enable us to develop an important analogy between the theory of the birational transformation of curves and surfaces. Let ABG be the triangle of reference of the curve, and ABCD the tetrahedron of reference of the surface ; then the following correspondence exists between the different elements of a curve and a surface, which is shown in the table on page 170. 233. The first theorem has alread}^ been proved, and the others may be established as follows. For brevity we shall write (ii) The first two portions are proved in § 165, where it is shown that when 2r tangents coincide in pairs, each pair being distinct, the constituents of the multiple points are S = \ — 7', K = r. In § 194 it is shown that when the tangent cone has 8 distinct nodal generators, the constituents of the multiple point on the surface are C = fji-8, B = 8. (iii) From § 165 and § 194 it follows that if a multiple point on a curve has r tangents, each of which consists of three coincident tangents, the constituents of the multiple point are S = \-2r, K = 2r, whilst, if the tangent cone at the multiple point on the surface has K cuspidal generators, its constituents are C = fi-2fc, B=2k. (iv) The first two portions follow from § 165 ; and the con- stituents of the multiple point on the curve are 8 = \-2r + \, /c = 2r-l, whilst the latter part follows from | 201, which shows that the constituents of the multiple point on the surface are G = ^JL-28^-l, 5 = 2S-1. 170 SINGULARITIES OF SURFACES m Q •p. A O 02 < ^ 1? Sh O .S ^ ID §1 . . - « cS O) r-H r-H 43 5 T3 O O ^ eg >j a; o 52 oo !=l 2 a t?^ =* <^ S o w m -^ -a a o -S 3 S t^ a &i o o (3^ -1-5 V. m O §2§ o S3 ■aU§ .2+3 a g "2 So OS o a 4> •3 G Hi 2 s o Si S c o 4j ii> "o S =« aj ■Is 3 a5 as ai o a o cS ^ -t^ a S ^ i=l o .a S 2 8 S^ ^ H O a> -a £>0 bo-5 a ^ ^ o d ^' o - <'^ p' O cS a . -^ ,;5 o ;a tn eg b ^ ^ rt *^ a a a -rt 3 cs '=' a o +3 a o ID 2~S a += o a v-^ a-^ ^ "73 += O ., O „ o -tj « C a> eg eg O " a += !-i ^ O C 2 «2 4)-;:^ +3 TO grd a O o eg a ors (D &0' IS O a i=l o Orn ^ o o ■— ' fH '^ Si 35 c« •'-' ^ rH 0) rG a O 3§J Ph o a 03 O O a '••' O -If w a.asH as ^oo ii i^S^ so cs"o a += o o S^ 2 o - += O Pl +3 o o) a ^ S ^ Ei^ ^ a a-o o a ' a a +3 &, 0^ Q o W a « +3 o P 03 % ^S3 '^ ^ c>r2-3 03 o --^ § 9 a § §k; "^ 03 O ^ i a rC3 a o O ff 02 . 03^ :i-j^ ^33 --- to ^ M ■=* -S g a O 02 o ■=*-! a "^ Z u '-I •^-^ ^^ •? o -^ I =« 03 --3 >5 O 03 cS jl/ Jh o o j3 I— I f-( CO y ^ -(J O 03 CO -tJ CM ?iH .g'H-t^ I -^^^^ ^ a 2 ° P c«.3 ^ . __, O 00 t» ^ <* " 2 2 ^a.2 !=! f= O r] ct c3 cS c6 32 ^ a 2 ? *-! p4-g O O CM bO 1 a 5~ ' ^ 33 ^ ce 03 o a . a g O cS --^ "^rS 2 > o-o a ^^ a a g o •" ce eg o fH Ph f-1 03 -73 . ^ a a 02 O S 03 J:3-a =5 ■^ a o eg o :g a ^ bO g-^ a cc S >. 0^ a *^ -S pa-H "^;3 o CO p2 .a +^ bc ^2 eg'" a o -J3 ^ «tH _03 ra o ;a += bfj bfl a '^ -a '+= I 02 03 ' 02 ■I~^ eg -r; .a I o o ?is ""2 a BIRATIONAL TRANSFORMATION 171 (v) By § 172 (i) it follows that if a multijDle point on a curve has s pairs of tacnodal branches and p — 2s ordinary branches, all of which are distinct, its constituents are S = \ + s, K = 0, and it follows from the theorem, § 24, that if the tangent cone at a multiple point on a surface has S nodal generators, all of which are lines of closest contact, the constituents are G=fi + 8, 5 = 0. (vi) Putting 2r for r in § 173, it follows that the constituents of the multiple point on the curve are S = X-2r + l, K = 2r. Let the tangent cone at the multiple point on the surface have S + 1 distinct nodal generators, and let one of them AB he a. line of closest contact. Then it is shown in § 201 that if the cone possesses S + 1 nodal generators, which move up to coincidence along a continuous curve, the total reduction of class is 2/x, + 2S + 1 ; whence s = 2S + l. Also it follows from §202 that if one of the generators before coincidence is a line of closest contact the con- stituents of the singularity are G = fM-2S+l, B=28, since we have shown that s = 28 + 1. (vii) The first two portions follow from § 174, which shows that the constituents of the multiple point on the curve are 8 = \ + s-2r + 2, K = 2r-2. To prove the third part, it follows from § 203 that if the tangent cone at the multiple point has r coincident nodal generators, all of which before coincidence are lines of closest contact, 2(7 + 35 = 2/i + 4r - 2, C + B = fjL + r. But if the tangent cone possesses S — r additional distinct nodal generators, all of which are lines of closest contact, their effect is to produce an additional reduction of class equal to 28 — 2r, and to add B — r double points to the constituents of the multiple point ; whence 2G + SB=2fi + 2B + 2r-2, C-\-B = ti+S, accordingly G = fji + S — 2r + 2, B=2r — 2. 172 SINGULARITIES OF SURFACES (viii) If in §173 we put r=l, we obtain the singularity in question, and its point constituents are 8=\ K = l, and it follows from § 202 that if the tangent cone at a multiple point possesses a cuspidal generator, which is a line of closest contact, the constituents are C=iJL, B = l. (ix) The first two portions are proved in 1 181 ; and the last two portions in § 231 (vi). The point constituents of the singulari- ties are S=X, k=1 for a curve ; and G = /x, B=l for a surface. 234. It may, at first sight, appear strange that two such different singularities as the ones discussed in (v) and (vi) of § 231 should have the same point constituents ; but the theory of curves supplies the explanation. The singularity corresponding to (v) in plane geometry is a multiple point consisting of one pair of tacnodal branches, one coincident ordinary branch, and p — 3 distinct ordinary branches ; and its constituents are given by the equations S = \, k = 1, t = 2, t = 0, whilst the one corresponding to (vi) is a multiple point consisting of a rhamphoid cusp and p — 2 distinct ordinary branches passing through it, and its con- stituents are given by the equation S = X, k = 1, t = 1, i=l. Both singularities are therefore mixed ones, whose point con- stituents are the same, but whose line constituents are different ; and from analogy we should anticipate that the singularities (v) and (vi) of §231 in solid geometry are mixed ones, whose con- stituents consist partly of point and partly of plane singularities ; but that the plane constituents of (v) and (vi) are different. CHAPTEH VI QUARTIC SURFACES 235. The class of a quartic surface may be any number lying between 3 and 36. In the latter case the surface is anautotomic and its equation contains 34 independent constants ; whilst in the former it is the reciprocal polar of a cubic surface, A quartic surface may also possess as many as 16 double points, which may be isolated or may coalesce so as to form a variety of compound point singularities as well as singular lines and curves. A com- plete investigation of quartic surfaces would require a separate treatise, and all that can be attempted in the present chapter is to give an account of some of the principal results, with references to the authorities where further information may be obtained. Nodal Quartics. 236. The theory of these surfaces has been worked out by Cayley* at considerable length. When the surface has not more than four nodes, these may be taken as the vertices of the tetra- hedron of reference, and the highest power of the coordinate a corresponding to any node A must be al The existence of each node involves one equation of condition ; but if the node is situated at a given point, three more equations are required to determine the point. Hence a given node involves four equations of condition ; accordingly if the surface has k given nodes, it cannot contain more than 34 — 4 8, this expression becomes negative, the explanation of which is that a quartic surface cannot possess as many as 9 nodes which are arbitrarily situated ; but the nodes must lie on one or more given surfaces called dianodal surfaces. We shall hereafter show * Proc. Lond. Math. Soc. vol. iii. pp. 19, 198, 234 ; and G. M. P. vol. vii. pp. 133, 256, 264. 174 QUARTIC SURFACES that 7 is the greatest number of arbitrarily situated nodes which a quartic surface can possess. 237. Five given nodes. Let A, B, C, Dhe four of the nodes, and let the fifth be at the point (/, g, h, k). Let P, Q, R, S, T be five quadric surfaces, each of which passes through the five nodes ; then the equation of the quartic is (P,Q,R,S,Tf = (1), for it obviously possesses nodes at the five given points, and since it contains 14 independent constants, it is the most general form of the required surface. The five quadrics may be taken to be P = ^(ky-h8), Q = ^{fy-ha), R^y(k^-gS), S = y{f^-g^\ T=fk^y-ghoLS, from which if we eliminate (a, /3, 7, B) it can be shown that there is one relation between the five quadrics, which is a cubic and not a quadric function. 238. Six given nodes. In the last article, the analysis may be simplified by writing a' = a/f, in which case the coordinates of the fifth node are (1, 1, 1, 1), and we shall take the coordinates of the sixth node to be (/, g, h, k). Let P, Q, R, S be four quadrics passing through the six points, then the equation of a quartic having these points as nodes is (P,Q,R,Sy = (2), but since this contains nine instead of ten constants, it is not the most general form of a sexnodal quartic. Let J be the Jacobian of the four quadrics, then it can be shown that the latter is a surface having the six points as nodes which is not included in (2). Hence the required equation is (P,Q,R,Sy+\J==0 (3), where A, is a constant. The four pairs of planes ^[a{h-k) + y{k-f) + B(f-h)]=0. y[a(k-g) + ^if-k) + 8(g-f)} = o\ B {a(g-h) + /3(h-f) + j{f-g)} = 0l a{^{k-h) + y(g-k) + 8{h-g)} = 0i pass through each of the six points ; but if we add together the equations of the second planes in each pair, the result vanishes, weddle's surface 175 which shows that the planes are not independent. We shall therefore take the first three pairs as the surfaces P, Q, R; and the surface S to be the cone S = g{k-h)yS + h(g-k)8^ + k{h-g)^y = (5) whose vertex is A, and which passes through the remaining five points. The Jacobian of P, Q, R, S will be found to be + {k - g) {ho? - M B^ + (h - f) {k/3^^ -gB^)ay + (g- h) {ka^ - /80 /3y + {k-f) (gy' - hfi^) aB . . .(6), from which it can be shown that J cannot be expressed as a quadric function of P, Q, R, S; also J has nodes at each of the points A, B, C, D and it can easily be shown that it has nodes at the two remaining points. Weddle's Surface. 239. Weddle showed* that the locus of the vertex of a quadric cone which passes through six given points is a quartic surface ; and we shall now show that this is the surface (6). The surface lP + mQ + nT+pS = (7), where (I, m, n, p) are arbitrary constants, represents any quadric surface passing through the six points. If this surface has a node, the coordinates of the latter are obtained by differentiating (7) with respect to (a, /3, 7, B) ; but since only three equations are necessary to determine a point, the elimination of (a, ^, dQjda + fdR/da + XdA/da =0 (12) weddle's surface 177 with three similar ones. The elimination of (a, /3, 7, 8) from (12) gives rise to a relation between the constants of (9) which reduces them to five ; but if (0, cf), i/r, X) be eliminated, we shall obtain the equation J(P,Q,E,A) = (13), where J is the Jacobian of P, Q, R, A. This is a surface of the sixth degree, and (12) shows that the eighth node may be any- where on this surface ; moreover the latter passes through the remaining seven nodes, and is therefore the dianodal surface. Since any given point on (13) requires two equations of condition for its determination, the equation of a quartic which possesses seven arbitrary nodes and an eighth one, which lies on the dianodal surface (13), contains three arbitrary constants. We have therefore shown that a quartic surface cannot possess more than seven conic nodes which are arbitrarily situated. If a quartic possesses more than this number, the nodes must lie on a certain surface (which need not be a proper one) called the dianodal surface. 243. The octonodal quartic (8) has been discussed by Cay ley* and is one of considerable importance. It will hereafter be shown that all quartic surfaces having a singular conic can be reduced to this form ; the equation also includes the reciprocals of the following surfaces, viz. parabolic ring n=6; elliptic ring n = 8; parallel surface of a paraboloid, and first negative pedal of an ellipsoid n = 10 ; centro-surface of an ellipsoid and parallel surface of an ellipsoid n = 12. Also the general torus, or surface generated by the revolution of a conic about any axis whatever. The proof of these theorems belongs to the theory of quadric surfaces rather than to that of quartics ; I shall therefore merely give the following investigation due to Cay ley f in order to illustrate the method to be employed. 244. (i) The centro-surface of an ellipsoid is the locus of the centres of principal curvature. Let P be any point on the surface, (x, y, z) the coordinates of either of the centres of principal * G. 31. P. vol. vn. p. 304; Quart. Jow: vol. x. p. 24; C. M. P. vol. vin. pp. 2 and 25. t "On the Centre -Surface of an Ellipsoid," C. M. P. vol. viii. p. 303; Trans. Camh. Phil. Soc. vol. xii. pp. 319—365. B. 12 178 QUARTIC SURFACES curvature corresponding to P, p either of the radii of curvature, ^ the perpendicular from the centre of the ellipsoid on to the tangent plane at P, then it is shown in Treatises on Quadric Surfaces* that (w, y, z) satisfy the equation a^x^ h^if cV , /to X + 77;^-rTv.+ 7:;^r^.= i (13a), where f = 'pp. Since the quantity ^ is a function of the position of P, the equation of the centro-surface is the envelope of (13 a) where |^ is a variable parameter, and its equation is therefore the discriminant of (13 A) regarded as a binary sextic (^, 1)^ — 0. But since the surface is the envelope of the ellipsoid (13 a), the re- ciprocal surface must be the envelope of the reciprocal ellipsoid {a? + ^fx'la? + (62 + |)^2/V&' + (c' + D'^V^' = ^, and since this is a quadratic equation in ^, its discriminant is (aV + 62^2 + C2^2 _ ^) (^2/^2 ^ ^2/^2 ^. ^l^^f^ ^ ^^2 + 2/2+ ^2)2^ which is of the form (8). (ii) The rings in question are the envelopes of a given sphere of constant radius c, whose centre moves on a conic section. Let ^ = 0, 2/^ = ^ax be the equations of a parabola; then the coordinates of any point on the curve are x = aQ'^, y = 2a0, z = 0. Hence the equation of the sphere is (x - ae-y + (y- 2aey + z^ = c", and the discriminant of this equation regarded as the binary quartic (6, 1)^ = gives that of the ring, which will be found to be of the sixth degree. The reciprocal polar is the envelope of the reciprocal of the sphere, whose equation can easily be shown to be the quadric ^2 = ^^2^ + 2a0y + cr, where k is the constant of reciprocation ; whence the equation of the reciprocal surface is (ay^ + k^xY = c^x" {x^ + 2/^ + z-), which is a quartic. Therefore the original surface is of the fourth class. 245. I shall not attempt to discuss the remaining cases in detail ; but there are a few points which require consideration. * Frost's Solid Geometry, vol. i. (1875), § 618. THE SYMMETROID 179 = 0, •(14), The equation a, h, g, I h, h, f, m g, f, c, n I, m, n, d where the letters represent arbitrary planes, is a quartic surface called the Symmetroid. It possesses 10 nodes, which lie on the cubic surfaces obtained by equating the minors of this determinant to zero. The Hessian of a cubic surface is a particular case of the symmetroid, and the existence of the nodes on the latter has been proved in § 60. It can also be shown that the vanishing of any of the four quantities a, h, c and d produces an additional node. When all four vanish, (14) assumes the form (lf)i + (mg)i + {nh)i=0 (15), which is a special case of a quartic with 14 nodes. The equation of a quartic surface having a conic node at A is a^Wg + 2aw3 + W4 = (16), and the tangent cone from A is the sextic cone ti/ = U2U4 (17). Now a proper sextic cone cannot possess more than 10 nodal generators ; if therefore a quartic surface possesses more than 11 conic nodes, the tangent cone (17) will degrade into an improper cone, and this fact has been made use of by Cay ley* for finding the equations of quartic surfaces with more than 11 nodes. Kummers Surface. 246. This surface has been so fully discussed in Mr Hudson's recent treatise f that only a slight sketch will be given. The equations of a quartic having a node at A and of the tangent cone from A, are given by (16) and (17) and the latter obviously touches the nodal cone u.^ along the lines of closest contact. The line joining A to any other node on (16) must be a nodal generator of (17); and since Rummer's surface possesses 16 nodes, the sextic cone (17) must possess 15 nodal generators and must therefore * Proc. Lond. Math. Soc. vol. iii. p. 234. t Kttmmer's Quartic Surface, Cambridge University Press. 12—2 180 QUARTIC SURFACES degrade into six planes. Each of these planes is intersected by the five other planes, and their five lines of intersection connect A with five nodes ; hence each of the six planes contains six nodes. But since each of the six planes forms a part of the improper tangent cone from A, each plane must touch the quartic along its curve of intersection and therefore the latter must be a conic twice repeated ; in other words, each of the six planes is a conic trope on which lie six nodes. Also since the surface is of the fourth class, it is its own reciprocal ; moreover the reciprocal polar of a conic node is a conic trope on the reciprocal surface, and since the original surface possesses 16 conic nodes, it must also possess the same number of conic tropes. 247. Kummer starts with the irrational equation {ot-u)^^^ + (a-vfy^ + {a-w)^B^=0 (18) or (a - uf /32 4- (a - vf r^''+ {a- wf 8" - 2 (a - ?;) (a - w) 7S -2(oi-w){a-u)B^-2{a-u)(a-v)/3y = (19), where u, v, w are any three planes passing through A. Equation (19) may be written in the form ^a^-25a + C = 0, where ^ = /32 + 72 + §2 _ 2^8 _ 2g^ - 2/37, B = ^hi + r^H + hhu - (w + w) 78 - {uj + u) S^-(ii + v) ^y, C = ^hi^ + rfv^ + Bhu'^ — 2vwyS — 2wuS/3 — 2uv^y ; hence ^ is a node. Writing (18) in the form it is obvious that the points of intersection of the eight triplets of planes viz. /3, 7, 8 ; l3,j,Z; jS, Y, S; jB, Y, Z; X, y, B; X, y, Z ; X, F, S ; X, F, Z are nodes ; but we shall show that (18) possesses altogether 13 nodes. Let cr = /3 + 7 -I- 8, 2 = jSit + 7^ + Bw, then ^=2(/32 + 72 + 8")-o-^ B = 2 (I3'u + y-'v + B-'w) - Xa, C=2 (/3V + y-v" + B'^vfi) - 1\ and if the above values oi A, B and G be substituted in the equation AC =^ B" it reduces to 2^yB {/3 (u — v) (w — ?/) + 7 (m — v){v — iv) + B(v — w) (vj — u)] = ; kummer's surface 181 hence the tangent cone from A consists of three planes and a proper cubic cone. The complete cone has accordingly 12 nodal generators, viz. the lines AB, AG and AD, and the three lines in which each of the planes ABC, AGD and ADB intersect the cubic cone ^l{v-w) + r^l{w-u) + ZI{u-v) = (20). The quartic therefore possesses 13 nodes ; but it will have 14 when (20) has a nodal generator; 15 when (20) consists of a quadric cone and a plane; and 16 when (20) consists of three planes. The condition that the plane ^jl + 7/m + hfu = should touch the cone J. = is that l + m + n = 0, and Kumraer takes u = l (^1^27 — m{ni2^), V = m (lil^B — n-ji^j^), 10 = 01 (mlW^2/S — ^1^27), where l + m + n = li+ m^ + n^ = h-\- ^1.2 + Wa = ; from which it can be easily shown that the equation of the tangent cone becomes Kl3yB (1311 + y/m + 8/71) (fi/h + y/m, + S/%) {^Jk + j/m, + S/n,) = 0, where K = ^llil^mmimonninz. 248. A particular case of Rummer's 16 nodal quartic surface is the Tetrahedroid, which can be projected into Fresnel's wave surface ^2^2 Jfy2 g2^2 J.2 _ g2 ^2 _ J2 ^,2 _ fZ The sections of the surface by each of the coordinate planes consist of a circlis and an ellipse, and if a>b> c the four points of intersection in the plane y = are real and give rise to four real conic nodes, which produce external conical refraction ; the eight points of intersection in the planes a? = 0, ^ = are imaginary, and give rise to eight imaginary conic nodes ; and the section by the plane at infinity consists of the factors ^^2 ^ y2 ^^2 ajj(J 0^2^.2 ^ l,2y2 _|. g2^2^ showing that there are four nodes on the imaginary circle at infinity *. * In 1871 Lord Eayleigh proposed a theory of double refraction, which is discussed in Chapter XV of my Physical Optics, in which the velocity of pro- pagation is determined by the equation ■{ A =0. ! ~ ^,2 ),2 ~ ™,2 ,.2 Prom this it follows that the pedal of Lord Eayleigh's surface is Fresnel's, and that 182 QUARTIC SURFACES The surface has also 16 conic tropes, four of which are real and the remaining 12 imaginary. The Hessian intersects the surface in the 16 circles of contact and the latter constitutes the spinodal curve. The flecnodal and bitangential curves do not appear to have been investigated. Quartics with Singular Lines. 249. The theory of surfaces with singular lines has already been given, and we shall now enquire what lines of this character a quartic surface can possess. Nodal line of the first kind. We have shown in §§ 43 and 37 that when the surface is of the nth. degree, the reduction of class i^ = 7w — 12, and that the number of pinch points is 2n — 4. Hence when n = 4, i2 = 16, m = 20 and the number of pinch points is four ; accordingly if A and B are two of them, the equation of the quartic is pa"^ + 2a/3v^ + q^Y + 2ai;3 + 2/3w., + tv, = (1). 250. The surface has 16 lines lying in it, all of which intersect the nodal line. The section of the surface by the plane S = ky consists of AB twice repeated and a conic ; and if the discriminant of this conic be equated to zero, it will furnish an equation of the 8th degree in k, which shows that there are eight planes in which the conic degrades into a pair of straight lines. There are thus 16 lines, which lie in pairs in eight planes passing through AB. See also §44. its reciprocal polar is the inverse of Fresnel's surface, and is of the 6th degree. The surface is therefore of the 6th class, and if a certain inequality existed between the optical constants, biaxal crystals would be capable of producing triple refraction. A principal section of the reciprocal surface consists of a circle and the inverse of an ellipse with respect to its centre, and since the last curve is a trinodal quartic, and therefore of the 6th class, the principal sections of Lord Eayleigh's surface consist of a circle and a sextic curve of the 4th class. The surface is therefore of the 8th degree. For a uniaxal crystal, this wave surface degrades into a sphere, and the reciprocal polar of the inverse of a spheroid with respect to its centre. The inverse of a spheroid can possess a pair of real tropes having real circles of con- tact, which reciprocate into a pair of real conic nodes having real nodal cones ; hence Lord Rayleigh's theory leads to the result that uniaxal crystals might not only possess tivo extraordinary rays as well as one ordinary ray, but might also produce external conical refraction. QUARTICS WITH A NODAL LINE 183 251. An arbitrary plane cuts the surface in a uninodal quartic, but a triple tangent plane cuts the surface in a pair of conies which pass through the point where AB intersects the plane, and intersect one another in three other points which are the points of contact of the plane. Let BCD be a triple tangent plane, then the equation of the surface must be of the form (aU+Pr)y' + {aV+PQ' + P'Q)yB + {aW+QQ')B"^ = 0...(2), where U, V, W are arbitrary planes, and P, P', &c. are planes passing through A ; for when a = 0, (2) becomes {Py + QB)(P'ry + Q'B) = = SS' (say) (3). Let AG he one of the 16 lines, ABD any one of the eight planes, D the remaining point where BP cuts S ; then lf=f^ + hB, P = \^ + vB, Q = G^ + Hy, F = \'/3 + fiy + vB, Q' = Q'^ + R'y + K'B, F= P(P'a + Q'^ + K'B) + F'Gt^ + %. Putting 8 = 0, (2) becomes ^{/a + X(X'/3 + /7)|=0 ....(4), which shows that J.0 is one of the lines lying in the plane ABC, whilst the other line is given by the remaining factor of (4). Let G' be the point where the last line cuts BG\ then it follows from (3) and (4) that G lies on the conic ^, and G' on the conic S'. Let P' be the remaining point where BP cuts the conic B' ; then putting 7 = in (2), it follows that the equation of the lines lying in the plane ABB is (Pa + Q^) {F'oL + G'^ + K'B) = 0, the first of which passes through the point D which lies on the conic 8, whilst the second passes through the point D' which lies on the conic 8'. Hence : — If BGP he any triple tangent plane, the section of the surface by it consists of two conies 8 and 8' ; also one of the lines in each of the eight planes intersects the conic 8, whilst the other intersects the conic 8'. Since the nodal tangent planes at B are \y + GB = and V7 + G'B = Q, it follows that : — The nodal tangent planes at B (ouch the two conies respectively. 252. The theorems of § 250 or § 44 show that eight tangent planes can be drawn to the quartic through the nodal line AB. Now a node diminishes the class of the surface, and therefore 184 QUARTIC SURFACES the number of tangent planes which can be drawn through an arbitrary straight line, by 2 ; hence the plane through the line and the node is an improper tangent plane, which is equivalent to two ordinary tangent planes. If therefore the surface possesses an isolated conic node, only six ordinary tangent planes can be drawn through AB, giving 12 ordinary lines, whilst the two remaining lines consist of a pair passing through AB and the node, each of which is equivalent to two ordinary lines. Since a binode reduces the class by 3, it follows that if the surface possessed an isolated binode, there would be only five planes and 10 ordinary lines, and each of the lines through AB and the binode would be equivalent to three ordinary lines. When the surface possesses four conic nodes, there are only eight lines, which consist of four pairs such that the lines belonging to each pair intersect at a conic node ; and in this case the equation of the surface may be expressed in the form {U,V,Wf = (5), where U, V, Ware three quadric surfaces, which possess a common straight line. The latter is the nodal line on the quartic, and the four distinct points in which the quadrics intersect are the four nodes on the quartic. 253. Since not more than eight tangent planes can be drawn through the nodal line AB, it follows that if the quartic has a fifth node it must lie in a plane through AB and one of the four other nodes ; for if not, five improper tangent planes, which are equivalent to ten ordinary tangent planes, could be drawn through AB to the surface, which is impossible. Now it follows from § 31 that if a surface of the nth degree possesses n — 1 conic nodes lying in the same straight line, the latter not only lies in the surface, but the tangent plane along it is a fixed instead of a torsal tangent plane ; accordingly if two conic nodes P and Q lie in the plane ABPQ, the point where the line PQ cuts AB is a third node on the section and therefore the plane must touch the quartic at every point on PQ. This result may be proved more simply for a quartic as follows. If there is a conic node at P, the section by the plane ABP must consist of AB twice repeated and two straight lines Pp, Pq ; but if there is another node at Q, Pp and Pq must pass through it and must therefore coincide. It therefore follows that: — Whe7i = 0, plijcker's surface 185 a quartic surface possesses a nodal line AB and eight conic nodes, the latter lie in pairs in four planes passing through AB ; also each of these planes touches the quartic along the line joining the pair of nodes lying in it. This surface is called PlilcJcer's Complex Surface, by whom it was studied in connection with the theory of the Line Complex*. 254. The equation of Pluckers surface may he expressed in the form a, h, g, a h, h, f ^ g, f, G, 1 oc, /3, 1, where a = b = h = { 7)" ; also these lines must be generators of the cone and the quadric aS + 72 = 0, 8^ + Swi + /3vi + V2 = 0, whence eliminating h, we obtain 7^-a7X + an/Svi + W2) = (27). • Equation (27) is that of a quartic cone whose vertex is D, on which DB is a triple generator; hence the constants must be determined so that (27) degrades into a pair of planes and a quadric cone which pass through DB ; but it will not be necessary to work out the necessary conditions, because (21) transforms (27) into itself, and therefore a pair of intersecting straight lines on the cubic which pass through the conic, transforms into a pair of intersecting straight lines on the quartic which pass through the nodal conic. This pair of lines on the quartic lie in the plane 8', for if we put 8' = in (24) it reduces to (27), We thus obtain the theorem : If a cubic surface pass through the conic /S = 0, aS + 7^ = 0, and is not touched at D by the plane a; equations (21) transform the cubic into a quartic having a nodal conic whose equations are a' = 0, ^'8' — y^ = 0. The 16 lines which intersect the conic through which the cubic passes transforvfi into the 16 lines on the quartic; the 10 lines which intersect the line AD on the cubic transform into 10 conies, which form five pairs lying in five planes passing through the point D' on the nodal conic; and the line AD on the cubic transforms into the point D' on the quartic. 278. The theory of quartics furnishes the following theorem with respect to cubics : Let a plane cut a cubic surface in any line AD and a conic S. Then (i) 16 li7ies pass through 8 ; (ii) each of these 16 lines is intersected by five others which pass through 8 and five which pass through AD ; (iii) of the first set no two lie in the same plane, and the same is true of the second set, but a plane can be drawn through any line of the first set and one line of the second set ; (iv) when two lines passing through 8 intersect, the residual intersection of the plane and the cubic is a line passing through AD. CUSPIDAL CONICS 199 Cuspidal Conies. 279. The equation of a quartic surface liaviog a nodal conic is given by (4) of § 266, where U=a.'^Uo + au^ + Uz] and the pinch points are the four points of intersection of the nodal conic with the quadric cone But if the conic is cuspidal every point must be a pinch point, which requires that U-^" — u^ — kD,, and (4) is reducible to the form a^u+U^ = (1), where u is r plane and U a quadric surface. The quadric U has tritactic contact with the surface along the cuspidal conic, and intersects it in a conic along which the quartic is touched by the plane u ; hence w is a conic trope. Let C and D be the points where the trope intersects the cuspidal conic ; then U, = XyB + fiS/3 + v^y J '^'^^' and (1) becomes a'{la + m^) + {aU,+ U^f^O (3). Since (3) may be written in the form I3V + I3un, + O2 = 0, it follows that G and D are tacnodal points. These are the only singular points on the cuspidal conic ; also the form of (3) shows that the cuspidal tangent planes envelope a quadric cone, whose vertex is the pole of a with respect to the quadric a Ui +' U2 = 0. 280. Every plane passing through the tangent to the cuspidal conic at the tacnodal points cuts the surface in a quartic curve having a tacnode cusp * thereat, the tangent at which is the tangent to the cuspidal conic. The equation of any plane through the tangent at C to the conic is \h-^v^ = koL (4), whence eliminating h between (3) and (4) we obtain a' (^a + myS) + [a [P^ + Qy + R {koc - v^)l\} + kay + /JL^ {ka - v^)/xy = . . .(.5), * The equations of a quartic curve having tacnodes &c. are given in § 179. 200 QUARTIC SURFACES which is the equation of a quartic curve having a tacnode cusp at G, and a = is the cuspidal tangent. 281. Every plane through the tangent to the conic of contact at the tacnodal points cuts the surface in a quartic curve having a rhamphoid cusp at this point, whose cuspidal tangent is the tangent to the conic. Let the conic trope cut AB in B', and let a' = Za + m/S ; then changing the tetrahedron to AB'GD, the equation of the surface DGCOIUGS a' (a - m/Sy + I {{a' - m^) U,-\- IU,Y = (6). The equation of any plane through the tangent at G to the conic of contact is l(\8 + v/3)-mQ^ = koi (7), whence proceeding as before the equation of the section will be found to be a' (a' - m^f + l\(Q + k) a 7 + La.'' + Ma'^ + N^'Y = 0. which is the equation of a quartic having a rhamphoid cusp at G. 282. The tangent planes at the tacnodal points cut the surface in two quartettes of straight lines. These eight straight lines may he divided into four pairs, such that each pair lies in a plane passing through both the tacnodal points. The tangent plane to the surface at G is Qa + v^ + X8 = (8), and the section of (3) by it consists of the four straight lines a'{la + m^) + {Pa^-(Roc + ,jil3)(Qa + v^)/xY = ...(9), and the section of the surface by the tangent plane at B, which is Roc + fi^ + \y = (10), consists of four straight lines which are the intersections of (9) and (10). 283. There are three Kummers cones, whose vertices lie on the line of intersection of the tangent planes to the quartic at the tacnodal -points; also each cone touches these planes and also passes through the cuspidal conic. Let G and D be the tacnodal points, A the vertex of one of Rummer's cones, then the equation of the quartic may be written in the form Za2(Za + 2w^) + (a?7i + n)2 = (11) CUSPIDAL CONICS 201 or a2 (Pa^ + 2?ma/3 - 2\a U, - 2\n - X'^a?) + (\a2 + aZ7a + n)2 = ...(12). Since A is the vertex of one of Kummer's cones, it follows from § 267 that l = \, f/'i = m/3, which reduces (12) to na?D.-{l 204 QUARTIC SURFACES Writing /S = X7 in (1), it becomes amp\ + q)ci + (f+g\)S] + 7 {a2 {FX' + h\ + G) + 2a8 (PX + Q) + S^} = 0, showing that the section consists of AD and a cubic, whose node is the point C where the section intersects BG. The condition that the cubic should degrade into a conic and a line through C is obtained by eliminating a and S between the coefficients of a^ and 7 in this equation, which furnishes a quartic equation for \. The surface possesses only four Kummer's cones. 291. A Nodal and a Cuspidal Line. When BG is a cuspidal line, every point on it must be a pinch point ; hence BG must be a generator of the quadric (3), the conditions for which are P=P^ G = Q% h = 2PQ (4), and (1) becomes a' [a'uo + otu, + (HB +h + gl3) 8] + 2RaryB^ + {Pa^ + Qay + rySy = ...(5). Putting a = 7 = 0, (3) now becomes {{g-2PR)^ + (H-R^)8] = 0, which shows that one of the pinch points coincides with B, so that there is only one distinct pinch point on BD. 292. The cuspidal line possesses two tacnodal points, through each of which a pair of straight lines can he drawn which lie in the surface. Let G be one of these points, ABG the tangent plane ; then Q = g* = 0, and (5) becomes a^ {aX + a (i?/3 + rh) + {HB +// + g^) h] + 2Pa7S^ + (Pa/3 + 78)= = . . .(6). To find the other point, let ^ = Icy, and change the tetrahedron to ABG'D; then the required condition is k (p - Pf+ 2EP'k) = 0, and since the root ^ = corresponds to C", there is one other tacnodal point. To prove the second part, put 8 = in ((>) and it reduces to a^(aX+i?a/3 + ^'/S-) = (7), TWO INTERSECTING DOUBLE LINES 205 which shows that two straight lines can be drawn through G, and similarly for the other tacnodal point. 293. There are four straight lines lying in the surface ivhich intersect the nodal line; and each of these lines intersects one of the lines which pass through the cuspidal line. Let AD be one of these lines ; then UQ = r = H =0, and (6) and (7) become a^ {pa^ +fjB + g^S) + 2RayB' + {Pa^ + 7S)" = . . .(8), and a'/S (pa + P-/3) = 0, which shows that the line AG through the point G intersects AD. Putting 7 = in (8), it becomes a'^(pa + g8 + P/3) = 0, which gives the other line lying in this plane ; and if in (8) we put j~ka, the discriminantal equation will furnish one other value of k, showing that there are two other lines. 294. The section of the surface by any plane through the point of intersection of the two lines, has a rhamphoid cusp thereat. The section by the plane 8 = ky is (Pal3 + kryj + a (a, ^, yf = 0, which shows that jB is a rhamphoid cusp. 295. Two Guspidal Lines. The line BD must also be a generator of (8), which involves the additional equations H = R^ g = 2PR; whence the equation of the surface becomes Oi'(uhio + oiu,+fy8) + (aU, + yBf = (9). Each line has one tacnodal point lying in it ; and if G and D be these points (.9) becomes a" {a'uo+pa^ + f (Qy + R8) a +fryS} + (aU, + ry8y = 0... (10), and the section of the surface by /3, which may be any plane through G and D, consists of a pair of conies which touch one another at these points. Also the tangent planes at G and D are Qa + S=0 and i?a + 7 = 0. 296. The tangent plane at each tacnodal point intersects the surface in a pair of straight lines ; each line of one system intersects one line of the other ; also the two points of intersection of the lines lie in a line passing through the point of intersection of the cuspidal lines. 206 QUARTIC SURFACES The section of the surface by the tangent planes at G and D are both represented by the equation (P/S - QRaf + (uo - QRf) a' +pci/3 = (11), which gives the projection on the plane ABC of the two lines ; and since they are identical, each line of one system intersects each line of the other system in a line passing through B. 297. The section of the surface by an arbitrary plane through the point of intersection of the cuspidal lines has an oscnode thereat. Let V=(Q + Rk)a + ky, and if the latter be taken as the side BG, F= 0. Similarly the section by the plane 7 consists of AB^ and the line 8/3 =GB, and if this be taken as the side AD, G = 0. We can therefore reduce (1) to the form ay% + ^B'w, = yB(Py'+QyB+RB') (2). The tangent plane at C, which may be any point on BG, is poi = PB, and the section of (2) by this plane consists of BG' and a conic passing through B. Hence BG is a singular line the tangent plane along which is fixed, and if this plane be taken as the plane a, P= 0. In like manner the tangent plane along AD is fixed in * CM. P. vol. V. pp. 168, 201 and vol. vi. p. 312 ; Phil. Trans. 1864 and 1869. t Mem. di Bologna, vol. viii. (1868). See also, Chasles, C. B. 1861 ; Rohn, Math. Ann. vols. xxiv. and xxviii. ; Segen, Crelle, cxii. QUARTIC SCROLLS 219 space, and if the plane be taken as the plane /3, R = 0. Accord- ingly (2) can be reduced to ay^Vi + ^S'w, = Qy'B' (3). The preceding argument shows that: — Through each of the four pinch points a singular line passes, the tangent plane along which is fixed in space. 327. The surface (3) gives rise to four species of scrolls. (i) dth species of Gayley ; 8th of Cremona. Since every plane through AB intersects the surface in AB^ and a line which passes through AB, all the generators pass through this line. Also writing the equation of the surface in the form ajB {py + qS) + ^Vs = v^ it follows that each of the tangent planes at the point A intersect the surface in a line passing through A , hence : — Through every point on the triple line three generators can he drawn, which lie in different planes 328. (ii) Zrd species of Cayley ; 9th of Cremona. When Q = 0, the line CD lies in the surface ; hence there is one generator which does not cut the triple line. The equation of the surface is now ary% + l3B''w, = (4), and the section of the surface by the plane /3 = \a is 7-V1 + XS^Wi = 0, and therefore consists of three straight lines which intersect on the triple line. Hence : — Through every point on the triple line three generators can be drawn, which lie in a plane passing through the generator which does not intersect the triple line. 329. (iii) 11th species of Cayley ; ^rd of Cremona. When ps = qr, the triple line becomes one of the second kind, since one of the tangent planes is fixed in space ; and (3) may be written (ay^ + 13^) (py + qB) = Qy^S' ....... .. .:...,. (5). There are only three distinct pinch points, since by § 216 one- pair coincide; and by forming the equation of the discriminantal surface as in § 216, it follows that the points A and B are the distinct pinch points, and that the other two coincide at a point 220 QUARTIC SURFACES A', such that q^a+p'^ = is the equation of the plane A'GD. Writing ^' = q^a-\-p^^, (5) becomes a (^7 - qh) (py + qSf + ^'B' {py + qB) = Qp'y'B' (6), which shows that A' is a pinch point. Writing (6) in the form ayB (py + qB) + ^yv^ = v^ , it follows that the planes B and py + qB intersect the surface in AB^ and two lines passing through A ; but that the fixed plane y intersects the surface in AB^, hence: — Through every point on the triple line only two distinct generators can he drawn, since the third one coincides with AB. 330. (iv) Qth species of Gayley ; 10th of Cremona. In this case the triple line is of the third species, one of the tangent planes being torsal and the other two fixed in space. There are only two distinct pinch points, by reason of the fact that they coincide in pairs. The equation of the surface is ay-'B + ^yB'^v, (7), where A and B are the pinch points. 331. (v) 10^^- species of Cayley ; 1st of Cremona. The quartic has a proper nodal twisted cubic, and its equation is {a, b, c,fg,h'^\,fji, vy = 0, where (X, /x, v) have the same meanings as in | 323, where this surface has been discussed. 332. (vi) 8th species of Cayley; 7th of Cremona. The scroll* ^(1, 3^) is a special case of the last species, since every generator intersects the cubic twice and also intersects a fixed straight line, whose equations may be taken to be la + m^ + ny -^-pB = 0, l'a + m'^-\-n'y+p'B = 0. Also if a = mn —m'n, f=lp' —I'p, h = nl' —n'l, g = 7np' — m'p, c = Im' — I'm, h = np — n'p, where a, b, c, f, g, h are the six coordinates of the line, the identical relation af+bg + ch = (8) Exists between the six coordinates. * The notation for scrolls will be explained in the next Chapter. QUARTIC SCROLLS 221 Let \ = fj, = v — be the equations of the cubic, where \ = /3S-7^ iii = /3y-0LS, v = ay-^'' (8a), then the parametric values of the coordinates of any point on the cubic are p, p6, p6^, p6\ and therefore the equations of the line through any two points 6 and on the cubic are a- p _ /3-pd _ j-pO^ _ B-pd' p — cr pd — a(f) pd'^ — cr^^ pd"^ — acf)^ ' which shows that the coordinates of any point on the line are given by equations of the form «=! + «, ^ = + (o(f), J = 6^ + &)^, where to is a variable parameter which depends upon the position of the point (a, j3, j, 8) on the line. Substituting these values in (8a) and rejecting the common factor Q)^{d — (f)Y, we obtain \: fi:v = d(t):-e-(j>:l. The conditions that the variable line should intersect the directrix line are I +m6 +n6'' +pe^ +q)(1 +m4> +w<^2 ^.^^s^^q, r + m'd + n'^2 + 'p'&' + « (r + m'<^ + n'^'' + p'<^^) = 0, eliminating w and dividing out by the common factor ^ — 0, we obtain c-h((f> + 0) +/{((/. + ey - (jiO] + a(f)e+g(f>e ( + e) + hj^'^e^ = o. tuting the values of 6 and ^ + ^ in terms of \, jj,, v this becomes 222 QUARTIC SURFACES 333. (vii) 7th species of Gayley ; '2nd of Cremona. The surface possesses a nodal conic and a nodal line which cuts the conic. Let J. 5 be the nodal line, (a, O) the nodal conic where n = X7S + /x8y8 + 1^/87 ; then the equation of the surface is a? (Ly-" + MyB + NB') + a(jjy + qS) ^ + D,' = .. .(11). Since the points C and D are any points on the nodal conic, let them be those in which the nodal tangent planes at A cut the conic; then L = N=0. The section of the surface by the plane AGJD now becomes yB{Ma^ + \0L{py + qS) + X'y8} = (12), which consists of a conic cutting the nodal conic at C and D, and of the straight lines A C and AD. These lines are obviously generators of the scroll. Returning to the more general equation (11), in which C is any point on the nodal conic, transfer the vertex A of the tetra- hedron to A', where ^ = ka is the equation of the plane A'CD. Then (11) becomes a% + awi [X7S + (fiS + vy) (ka + 13')] + {\yS + ...^ = 0, and the condition that the line A'G should lie in the surface is L + pvk + k'^v^ = 0, which shows that there are two points A', corresponding to any point G on the nodal conic, such that two lines GA', GA" lie in the quartic surface. Hence the latter is a scroll. Also if A be one of these points, Z = and (11) becomes a^S (ilf7 + i\^S) + a^ifl + O^ = (13), the section of which by the plane S is 72/8(^a + i//3) = 0. Hence : — Through every point on the nodal conic tiuo straight lines can he drawn lying in the surface, both of which inteisect the nodal line ; also the two points of intersection have one conmion tangent plane in which both the ttuo lines lie. The section of (13) by an arbitrary plane through AG consists of this line and a cubic curve; but from the first portion of this article, it is obvious that there is a certain position of this QUARTIC SCROLLS 223 plane such that the section consists of the line AG, another line AD' intersecting the nodal conic in D', and a conic cutting the nodal conic in G and D\ Hence : — The scroll may be generated by a line which intersects (i) the line AB ; (ii) a conic passing through B and lying in the plane BGD ; (iii) a conic lying in the plane AGD, which intersects the first conic in G and D, but which does not pass through A. 334. (viii) llth species of Gayley ; Uh of Gremona. Equation (11) shows that both nodal tangent planes are torsal, hence the nodal line is of the first kind ; but if iV^= the tangent plane 7 is fixed and the nodal line becomes one of the second kind. The section of the surface by the plane 7 consists of AB^ and a line through the point B, where the nodal line intersects the nodal conic. In both cases an arbitrary section of the surface by a plane through the point B has a tacnode thereat; and therefore the section belongs to the same species of curves as the conchoid of Nicomedes, which possesses a node and a tacnode. 335. (ix) 2nd species of Gayley ; 5th of Gremona. In this case the nodal conic degrades into two straight lines ; hence the nodal curve consists of three straight lines, one of which cuts the remaining two which lie in different planes. Hence if AB, BD and DG be the three lines, the equation of the surface is a?h (3Iy + M) + a(py + qS) jSy + jSy- = (14). 336. (x) 4 Let Vn be what Vn becomes when 7 = A, 8 = k ; then since P lies on the curve (2) g'^Vo +y-^Wi' + g''-^' + Vr,' = 0, which by the last two of (4) becomes (\gyvo + (\gy'-'v, + Vn = (9). Eliminating \g between (8) and (9) we obtain y8»SX + (a + S) /S"-^S"-iWi + {a + 8Yvn = 0... (10), which is the required equation of the scroll. If the tetrahedron be changed to ABGD', so that the directing line GV lies in the plane BCD', (10) becomes /3«8% + a'/3"-i 8^-1^1 + a'^Vn = (10 a). 359. To find the equation of the scroll S(l, 1, n). Since a hyperboloid can be described through any two straight lines, the doubly directing line may be supposed to consist of two generators belonging to the same system of a hyperboloid, which are indefinitely close together. Let AB he the doubly directing line. A' any point on AB; (2) the directing curve n\ also let the tangent plane to the hyperboloid at A' intersect (2) in P ; then 234 SCROLLS A'F is the line which generates the scroll. The equation of the hyperboloid may be taken to be ciy = ^S (11), and if 6a = ^ be the plane A'CD, the tangent plane to (11) at ^'is 7 = ^S (12), and the coordinates of P are (0, g, 6k, k). Hence the equations of AT are ~^^7=^^^ek'^k^^ ^^^^' From the first two of (13) we deduce (5), whence eliminating 6 by (12) we obtain Xg = (^S-ay)/S. Accordingly from (9) we get (^8-ayTvo + (^B-ayy-'Bv,+ 8''Vn = (14), which is the required equation of the scroll. 360. Equations (10) and (14) furnish a method of classifying the scrolls 8(1, 1, n) and ^(1, 1, n), which depends on the character of the curve n and not on the degree of the scroll. Let this curve have a multiple point of order p Q,t B and of order q at C. Let p have any value from to ?i— 1, and q any value from to r, where r is a number whose limiting value is obtained from the condition that the curve n is always a proper curve. Then by considering all possible curves of given degree subject to these conditions, we obtain the equations of all possible scrolls generated by them. 361. Cubic Scrolls, (i) Let the nodal line AB he the curve I, and the line yS = 0, a + 3 = or CD' be the line m ; and let the curve w be a plane cubic whose node is at B. Then in the formulae of § 847, we must put ? = m = 1, w = 3, g = 2, r = 0, in which case the lines AB, CD' and the plane cubic will be multiple lines of orders ^—p, 1, 1 ; but since AB is a nodal line ^ = 1, and conse- quently the line CD' must intersect the cubic curve in one point. Let C be this point, then the equation of the cubic curve is /3v2 + Bw„ = 0, and by (10) that of the cubic scroll is QUARTIC SCROLLS 235 which is the equation of a cubic surface having a nodal line of the first kind, and is of the form 5(1, 1, 3). (ii) In the case of the cubic scroll *S^ (1, 1, 3) the line CD' becomes one indefinitely close to AB, and therefore BG must be the tangent at B to the section by the plane a, the equation of which is therefore ^hv, + ^3 = 0. Accordingly by (14) the equation of the scroll is (/3S - a7) ^1 + ^3 = 0, which is the equation of a cubic scroll having a nodal line of the second kind. Quartic Scrolls. 362. We have already considered the different species of quartic scrolls, and we shall now explain Cayley's method of generating them. There are three species of the form S{1, 1, 4) and three of the form S(l, 1, 4). 1st species. This scroll is of the species S(l, 1, 4); AB and CD' are nodal generators, and therefore the section by the plane a is ^% + /SSWa + ^^2 = 0, and therefore by (10 a) the equation of the scroll is a'Va + a'/Swa + ^% = 0. 2nd species. Let the generating curve have a tacnode at B and a node at G; let BD be the tacnodal tangent and GD one of the nodal tangents at G. Then the equation of the section is and that of the scroll is I3y + a'/3yv^ + a'^Sw, = 0, which is of the same form as (14) of § 335. Srd species. Let the generating curve have a triple point at B and pass through G. Then its equation is ^Vs + 8w3 = 0, and that of the scroll is fivs + afws = 0, which can be reduced to (4) of § 328 by taking A and B as two of the pinch points. 236 SCROLLS 363. The next three species are of the form S(l, 1, 4). 4l{l + m) or n>l + m, a proper curve of degree l + ni cannot be drawn through the group p + q. Hence the only curve of degree I + m which can be drawn through the group p + q is the improper curve GiG„i= ; accord- ingly in this case also [p +q] = 0. II. The Multiplication Theorem. If the group p has a zero residual, then np where n is any positive integer has also a zero residual. This at once follows as a corollary of the addition theorem ; but there is no division theorem, that is to say if np has a zero residual it does not follow that sp, where s is any factor of n, also has a zero residual. This may be proved as follows. Let a proper conic touch a proper cubic at A, B and G; then [2 J. + 25 + 2CJ = 0; but A + B+ G cannot have a zero residual unless the three points lie in the same straight line, which is contrary to the hypothesis that the conic is a proper one. III. The Subtraction Theorem. Ifp + q and p be two point groups on a curve Gn, each of ivhich has a zero residual, then q has a zero residual. The groups p + q and p are the complete intersections of G^ with two curves (7z+„i and Gi respectively ; hence ii I + m = n, the equation Gi+mSo + GnSi+m-n + GiS^ = 240 THEORY OF RESIDUATION represents some curve which passes through the group p. But by hypothesis it must be possible to determine the arbitrary constants so that this curve passes through the group 'p ■¥ q] hence it must be possible to determine a curve ^S^^ which intersects G^ in the group q and nowhere else. Let l + m 1(1 + m) or n>l + m, the curve C^+,^ must be an improper curve consisting of two curves Gi and Gm, of which the latter intersects Gn in the group q and nowhere else. Hence in both cases q has a zero residual. IV. The Theorem of Eesiduation. If two point groups p and q have a common residual, then any residual of p is a residual of q. Let r be the common residual of p and q ; and let s be a re- sidual oip. Then by hypothesis [p+r] = 0, [g-i-r] = 0, [_p+s] = 0; adding the second and third we obtain [p + q + r + s] = Q, and subtracting the first, we get [^ + s] = 0, which shows that s is a residual of q. In the preceding theory we have expressly assumed that none of the curves pass through a multiple point on any other curve, so that all the points are ordinary points. The case of a node will be discussed later on. The theory is also subject to certain exceptions*, when the points composing any group such as p are so situated that a curve of lower degree than I can be described through them. For example, a cubic is the curve of lowest degree which can be described through six arbitrary points on a given curve ; but if the six points were so situated that a conic could be described through them, an exceptional case would arise. * Bacharach, Math. Annalen, vol. xxvi. p. 275 ; Cayley, C. M. P. vol. xii. p. 500. An exceptional case occurs in a theorem proved by myself, Quart. Jour. vol. xxxvi, pp. 50 and 51. EXAMPLES OF RESIDUATION • 241 368. We shall now give some examples. (i) If a straight line intersects a curve of the nth degree in n ordinary 'points, the tangents at these points intersect the curve in n {n —2) points which lie on a curve of degree n — 2, which is called the satellite curve. Let the straight line intersect the curve in the group P, and let the tangents at P intersect it in a group Q. Then [P] = and therefore [2P] = 0. The tangents form an improper curve of degree n, which inter- sects the primitive curve in the groups 2P and Q ; hence [2P+Q]=0. Accordingly by the subtraction theorem ' [^] = 0. Since the group Q contains 71^ -2n^n (n — 2) points, a curve of degree n — 2 can be drawn through them. Also every curve of degree n can be expressed in the form ft ^n—2 "I" tit^ • • • fji ^^ '-'j where a is the line, ti ... tn the n tangents at the points where it cuts the curve, and Sn-2 the satellite curve. (ii) If from any point 0, n{n—l) tangents be drawn to an anautotomic curve, the points where the tangents intersect the curve lie on one of degree {n — 1) (n — 2). The n{n — l) points of contact form a group P, which is the complete intersection of the curve and its first polar with respect to ; hence [P] = and therefore [2P] = 0. The tangents form an improper curve of degree n(n— 1), which touches the curve at the group 2P and intersects it at a group Q consisting of n (n — l) {n — 2) points ; hence [2P + Q] = 0, whence by the subtraction theorem [Q] = 0; hence the group Q is the complete intersection of the primitive curve with one of degree {n — l){n — 2). B. 16 242 THEORY OF RESIDUATION (iii) // six of the points of intersection of a cubic and a quartic lie on a conic, the remaining six points of intersection lie on another conic; also the four remaining points where the two conies intersect the quartic are collinear. Let 6 and 6' be the two groups of six points in which the cubic intersects the quartic, and let the group 6 lie on a conic. Then the conic will intersect the quartic in two points 2, and the straight line through 2 will intersect the quartic in two other points 2'. Hence [6 + 6'J=0, [6 + 2] = 0, [2 + 2'] = 0. Adding the first and third and subtracting the second, we obtain [6' + 2'] = 0, which shows that the eight points 6' + 2' lie on a conic. Let a straight line intersect a quartic in the four points S, S', T, T' ; then since the straight line repeated three times forms an improper cubic, it follows that if a conic can be described oscu- lating the quartic at 8 and S', another conic can be described which osculates the quartic at T and T'. (iv) A cubic can be drawn through the six points, where the stationary tangents of a trinodal quartic intersect the curve, which osculates the quartic at the T points*. Let / denote the six points of inflexion, and /the points where the tangents at the former points intersect the quartic. Then since the six stationary tangents form an improper sextic, [3/+J] = 0. It is a known theorem that if S and T denote the 8 and T points, the eight points I and 8 lie on a conic ; hence [/+/Sf] = 0; also[>Sf+r] = 0; accordingly [37 + 3^ = 0, [3>Sf + 3r]=0. Subtracting the first and fourth and adding the fifth, we obtain [J + 3T] = 0. * See Appendix I in which the S, T and Q points of plane quartic curves are explained. The S points are those denoted by P and Q in Cubic and Quartic Curves, § 19.3 (iv) ; and the two remaining points in which the line joining the S points cuts the quartic are called the T points. If the tangents at a node intersect the quartic in D, D', the hne DD' cuts it in two other points called the Q points. See also Basset, Amer. Jour. vol. xxvi. p. 169. NODOTANGENTIAL CUHVES 243 which shows that the six points /and the two T points three times repeated lie on a cubic. (v) A conic can he described through the six Q points and the two T points of a trinodal quartic. Let D denote the six points where the nodal tangents intersect the quartic; then since the three straight lines passing through each pair of D points and the corresponding pair of Q points form an improper cubic [i) + Q] = 0. It is a known theorem that a conic can be described through the six D points and the two S points, whence [D + S] = 0, also [8-{-T] = 0; accordingly [Q + J'] = 0, All the preceding theorems can be proved by trilinear coordi- nates ; and (iv) and (v) by the parametric methods applicable to trinodal quartics*. 369. We shall now consider how these results are modified when a nodef forms part of the group; and we shall confine our attention to the case in which two curves have the same node and the same nodal tangents, and such curves will be called nodo- tangential curves. We shall define a cluster of points to be any special arrangement of points indefinitely close together. If a nodal curve he cut hy tivo nodotangential curves in two groups of ordinary points p and p + r, then r has a zero residual. Let the curve Gn have a node at A, and let the nodotangential curve Gi cut G^ in a group of ordinary points of degree p. Then Gn = a^'-^Mg + cC^~hi.i + . . . Uy (3), Gi = aJ'-^iL^ + a^-%3 + . . . w,; also let 8r = Ir'Wh a'-* Sr = %'Wk(f-^ (4), and consider the curve ^l+m = ^11 ^^ l+m—n + ^l ^m = ^ (5), where I + m ^7i. Multiplying out, we obtain 2z+r«=a^+"*-n< + w;o)w2+ (6), * See E. A. Eoberts, Proc. Lond. Math. Soc. vol. xvr. p. 44. t Basset, Quart. Jour. vol. xxxvi. p. 43. 16—2 244 THEORY OF RESIDUATION which shows that Xi+m is a nodotangential curve ; hence Cn, Gi and %i+m each pass through the same cluster A of six points at the node. Let p + r be the number of ordinary points in which S^+m intersects Gn ', then p = ln — A, p + r = (I + m)n — A, whence r = mn. Now by hypothesis [p-\-A'\ = and ['p-[-r -\- A'] = 0, and equation (5) shows that it is possible to determine a curve S^^ Avhich intersects G^ in a group of 7mi ordinary points and nowhere else ; hence r has a zero residual. When l^-m< n, "Xi+m is of the form GiSm, and the same result follows. 370. Let two nodotangential curves intersect a nodal curve in two groups of ordinary points p and q; then p and q are coresidual. By hypothesis we have the following equations [A+p] = 0, [^+^]=0 (7). Draw any other nodotangential curve cutting the primitive curve in the groups p and q and in a further group of ordinary points r ; then [A+p + q + r] = (8). The theorem of the last article shows that we may subtract (7) from (8) in the same way as if they were groups of ordinary points; we thus obtain the two equations [p + r] = 0, [q + r] = 0, which show that p and q are coresidual. We may therefore apply the theory of residuation to nodotangential curves in the same way as to groups composed of ordinary points ; also the theory applies to nodotangential curves having any number of nodes, and is also true when the double points are biflecnodes. 371. Since a curve, which has none but ordinary nodes, its Hessian and its nodal tangents form a nodotangential system, it follows that : — (i) On a curve ivhich has none but ordinary nodes, the points of inflexion and the points ivhere the nodal tangents cut the curve forin, a pair of coresidual point groups*. * Richmond, Proc. Lond. Math. Soc. vol. xxxiii. p. 218. EXAMPLES OF RESIDUATION 245 Also since the first polar with respect to the node of a nninodal curve (including the case of a biflecnode) is a nodotangential curve, it follows that : — (ii) The points of contact of the tangents drawn from the node of any uninodal curve, the points of inflexion and the points luhere the nodal tangents intersect the curve form a coresidual system. Let /, D and E denote the number of points of inflexion, the number of points where the nodal tangents intersect the curve, and the number of points of contact of the tangents drawn from the node. Then for a uninodal quartic I = 18, D = 2, E = 6 ; also the two Q points are a residual of D. Hence : — (iii) The 18 points of inflexion of a uninodal quartic lie on a quintic, which passes through the Q points. It will hereafter be shown that every quintic which passes through the points of inflexion passes through the Q points ; hence if C^ be the quartic and Og the quintic, there is a triply- infinite system of such quintics which are determined by the equation Cg + (la + m/3 + nj) G^ = 0. When the node becomes a biflecnode the D points coincide with the node, hence : — (iv) The 16 points of inflexion of a unihiflecnodal quartic lie on a quartic. For a binodal quartic, 7=12, i) = 4; also the D points lie on a conic passing through the nodes ; but if the conic degrades into a straight line passing through the D points and one passing through the nodes, [Z)] = 0, hence [/] = 0. Accordingly : — (v) If the four points, where the nodal tangents of a binodal quartic ititersect the curve are collinear, the 12 points of inflexion lie on a cubic. When the nodes are biflecnodes, this becomes : — (vi) The eight points of inflexion of a quartic with two biflec- nodes lie on a conic*. Assuming the theorem of Cubic and Quartic Curves, § 194, we obtain : — * The expression for the radius of curvature of a Cassinian, see Cubic and Quartic Curves, § 251, combined with the theory of projection, furnishes a direct proof of this theorem. 246 THEORY OF RESIDUATION (vii) The six points of injiexion of a trinodal quartic lie on a conio ivhich passes through the S points. For a uninodal quartic E = 6; hence : — (viii) The six points, tuhere the tangents drawn from the node of a uninodal quartic touch the curve, lie on a conic passing through the Q points. 372. It is a well known theorem that every cubic which passes through eight of the nine points of intersection of two given cubics passes through the remaining one ; and we shall now prove a more general theorem. If I be any integer not less than n — 2, any curve of degree I which passes through ln-^{n-l){n-2) of the points of intersection of two given curves Ci and Cn passes through all the rest. Let the points of intersection of the curves Gi and (7„ be divided into two groups p and In—p; then since the coordinates of the points of the group In — p satisfy the equation C^ = 0, and consequently satisfy In—p equations of condition, the number of available constants, which any curve of degree I passing through this group contains, is ^l(l + S)-ln+p (9). The equation of any curve of degree I which passes through the points of intersection of Ci and Gn is Gi + GnSi^n = 0, provided I = n, and it therefore contains l{l-n + l){l-n + 2) (10) available constants; and if the curve through the group In—p passes through p, the expressions (9) and (10) must be equal ; whence ^l (I + 3) - In +p = ^(l -n+l){l-n + 2), giving p = ^{n — 1) {n — 2). When l = n — l or n — 2 the theorem is also true ; since in this case ln — p = ^l{l + S). 373. Let ^ = 5, ?i = 4 ; then p = 3 ; hence every quintic which passes through 17 of the points of intersection of a quartic and BACHA race's THEOREM 247 a given quintic passes through all the rest. Accordingly every quintic which passes through the 18 points of inflexion of a uninodal quartic passes through the Q points. 374. The theorem of § 372 is due to Cayley* ; but Bacharachf has pointed out an important exception to it. Let n>S; then the value of p may be written in the form p = ^n (n — 3) + 1. Now it is not in general possible to describe a curve of degree n — 3 through the group of points p ; but whenever this can be done, Cayley 's theorem is not true. This may be proved as follows. Through the group p describe a curve Gns, which cuts the curve Cn in a group of s ordinary points, where s = ^n (n — S) — 1 ; and through the group s describe another curve 0'„_3, which cuts On in a group of q points, where q=p. Then s + q = n(n — d). Now the curve wC^ n-3 + Gn^i-s — is one of degree l + 7i — S which passes through the group In and also through the group q + s; hence [ln + q + s] = 0. But since the group ^ + s is the complete intersection of Cn and Gn-3, it follows that [p + s] = 0, whence [In — p + q] = 0, which shows that q is a, residual of the group In— p. 375. Every curve of degree m, ivhich passes through In — ^{l + n — m—l){l + n — m — 2) of the points of intersection of two curves Gi and G^ passes through all the rest, provided m~l and 7n ~ I + n — 2. Any curve of degree m which passes through In—p of the points of intersection of Gi and Cn contains ^m {m + S) — In + p (11) * C. M. p. vol. I. p. 25. t Math. Ann. vol. xxvr. p. 275. 248 THEOEY OF RESIDUATION constants ; but the equation of a curve of degree m which passes through the complete intersection of Ci and Cn is Sm-l^l + Gn^mr-n = 0, and the number of available constants which it contains is |(7?i -l+l)(m-l+2) + i(m -n + 1) {m - ?i + 2) - l...(12), and if the curve which passes through the group In—p also passes through p, the expressions (11) and (12) must be equal, which gives 'p = \{l + n — m — 1) {I -\- n — m — 2). In this theorem various exceptional cases arise, which have been discussed by Bacharach in the paper referred to. A corresponding theory exists with respect to the intersections of surfaces, a brief account of which together with references to the original authorities will be found in Pascal's Repertorio di Mate- 7natiche Superiori, vol. ii. pp. 297 — 303. Theory of Residuation of Surfaces. 376. When we attempt to apply this theory to surfaces, we are at once confronted with a difficulty. Let the primitive surface Cn be intersected by another surface Ci in a multipartite curve of degree In, which does not pass through any singular points or curves on either surface. This curve may be divided into two groups of curves p and r ; but if the curve of intersection of the two surfaces is a proper curve, it will be impossible to describe an algebraic surface through the group r which does not pass through the group p. It is of course possible to perform the mechanical operation of describing a surface, such as a cone, whose vertex is any arbitrary point and whose generators pass through the group of curves r ; but if such a surface could be represented by an equation, the latter would be a transcendental and not an algebraic one, and the ordinary theory of algebraic surfaces would not apply. It is therefore necessary to suppose that the curve of intersection of the two surfaces is a compound one, consisting of two complete curves of degrees p and r, in which case it will be possible to describe an algebraic surface through r which does not pass through p. We may therefore extend the theory of the residuation of plane curves to surfaces in the following manner. Let the curve THEORY OF RESIDUA TION OF SURFACES 249 of intersection of two algebraic surfaces be a compound curve consisting of two complete curves of degrees p and r; then the curve p will be called a residual of r and vice versa. Hence two curves p and r on the primitive surface are said to be residual to one another, whenever it is possible to draw another surface through them which does not intersect the primitive surface else- where. Also the compound curve ^ + r is said to have a zero residual, which is expressed by the symbolic equation [p+r] = (1). Through the curve r draw a surface G^ which intersects the primitive surface in another curve of degree q, where mn = q + r; then the curves p and q have a common residual r, and are called coresidual curves, and this is expressed by the symbolic equation [p-q] = ...(2). 377. The theory of residuation of surfaces, like that of plane curves, depends on three subsidiary theorems, which may be respectively called the addition theorem, the multiplication theorem and the suhtraction theorem. The Addition Theorem. Ifp and q he two curves on a surface, each of which has a zero residual, then the compound curve p + q has also a zero residual. Since [p] = 0, this curve must be the complete intersection of a surface Gi with the primitive surface (7„ ; and for the same reason the curve q must be the complete intersection of a surface (7,„ with Gn. Let I -l-m^n; then the surface GnSi+m-7i + Gi Gjtn =0 (3) obviously passes through the compound curve p-\-q\ also since the degree of the surface is I + m, whilst that of the curve p + q\B n{l + m), the surface (3) cannot intersect G^ elsewhere. When l + m< n, the only surface of degree I + m which can be drawn through the two curves is the improper surface GiG^. The Multiplication Theorem. Ifp has a zero residual, then np, where n is any positive integer, has also a zero residual. This follows at once as a corollary of the addition theorem. The Subtraction Theorem. If p + q and p be two curves on a surface, each of which has a zero residual, then q has a zero residual. 250 THEORY OF RESTDUATION Let the curves p + q and p be the complete intersections of Gn with two surfaces C;+^ and Gi ; and let l + m = n. Then the surface (7;+^ must be of the form Gi+m^o + GnSi^m-n + Gi8m = 0, for this surface is of degree l + m and passes through the curve p, whose degree is In, which lies in the three surfaces C;+^, Gn and Gi. Now the curve q lies in the surfaces Gi^^i and (7„, and there- fore it must be possible to determine a surface 8m which intersects Gn in the curve q and nowhere else ; hence q has a zero residual. When l + m Gn and Gi, it follows that the intersection of G^ and Gi gives [^+P] = (11). The intersection of S;+m and Gi gives [^ + P + Q] = (12), and the intersection of S'l+m-n and Gi gives [Q] = (13), which proves the theorem. 382. Let AB be an oy^dinary line on the primitive surface S, and a midtiple line of order p on another surface S'; and let S and S' intersect in a residual cm've P. Draw a second surface S" which has p-tactic contact with 8 at every point on AB, and intersects S in a residual curve Q. Then P and Q are coresidual curves on S. The three surfaces intersect one another along AB in a cluster A of lines, which consist of AB repeated p times. Hence [A + P]=0, [A + Q] = (14). Through the curves P and Q draw any other surface which has ^-tactic contact with 8 at every point on AB, and intersects 8- in a residual curve R ; then [A + P + Q+R] = 0, whence by the last article [P + R] = 0, [Q + R] = 0, which shows that P and Q have an ordinary residual R, and are therefore coresidual curves. 383. The theory is of a similar character when G^i is taken as the primitive surface. The form of (10) shows that the surfaces S and Gn intersect in a curve of degree nm=Q', which is the com- plete curve of intersection of Gn and 8m ; and in another curve 254 THEORY OF RESIDUATION which is the complete curve of intersection of Gi and 0^. Hence as before we obtain [J.+P] = 0. Also the intersection of S;+m and Gn gives [^+P+Q'] = 0, and that of G^ and Sm gives . [Q'] = o. By means of these equations we can prove, as in § 882, the theorem : — Let AB he a multiple line of order p on the primitive surface S, and an ordinary line on another surface S' ; and let S and S' intersect in a residual curve P. Draw a second surface S" which has p-tactic contact with S' at every point on AB, and intersects S in a residual curve Q. Then P and Q are coresidual curves on S. 384. If a twisted curve is the partial intersection of two surfaces Gi and Gn, where I = n, which are such that AB is a midtiple line of order p on Gi and an ordinary line on Gn ; then the curve is the partial intersection Gn with another surface Si, which has p-tactic contact with Gn at every point on AB. Consider the surface GiSo + GnSi-n =Si=0. Since Si-n is a general quaternary quantic of degree I — n, the highest powers of a and /3 in Si are the (l — l)th powers ; hence AB is an ordinary line on Si; also the form of Si shows that it intersects Gn in the complete curve of intersection of Gi and Gn and nowhere else. The curve of intersection of Si and (7„ must therefore consist of the above mentioned curve and the line AD repeated p times ; and since AB is an ordinary line on both surfaces, they must have jj-tactic contact at every point on AB. This result is of importance in the classification of twisted curves. Also it can be proved in the same manner that : — If a twisted curve is the partial intersection of two surfaces Gi and Gn, where l^n, which are such that (i) AB is a multiple line of order p -^ 1 on Gi, and (ii) the tangent plane to Gn along AB is one of the tangent jjlanes to Gi along the same line ; then the curve is the 2iartial intersection of Gn with another surface Si luJiich has p-tactic contact with Gn at every point on AB. TWISTED SEXTIC CURVES 255 385. A twisted curve which is the partial intersection of a quadric surface and a surface C^, where the residual intersection consists of p distinct lines lying in different planes, is the partial intersection of the quadric with another surface Sn, which intersect in a common line, which is an ordinary line on the quadric and a multiple line of 07'der p on Sn- Let the surfaces be (Pa + Q^)ry = (Ra + Si3)B (15), ay = ^S (16), where P, Q, R, S are quaternary quantics of degree n — 2 ; whence eliminating (7, 8) it follows that (15) may be replaced by Ra' + (S-P)a0-Q^' = O (17), on which CD is a nodal line. Let another generator (u, v) of the same system be common to (15) and (16); then (16) must be expressible in the form au = j3v (18), and P, Q, R, S must be linear functions of (u, v); whence eliminating (u, v) between (15) and (18) we obtain an equation of the form on which CD is a triple line. Proceeding in this way, we obtain the theorem. Twisted Sextic Curves'^. 386. There are five primary species of twisted sextic curves, I. The complete intersection of a quadric and a cubic surface. II. The partial intersection of two cubic surfaces, when the residual intersection consists of a twisted cubic curve. Their equations may be expressed by means of the system of deter- minants A, A', u, u' B, B', V, v' C, G\ w, w where all the quantities represent planes. * Clebsch, Crelle, vol. lxiii. ; Nother, Crelle, vol. xciii. ; Pascal, Lincei, 1893, p. 120. Septimic curves have been discussed by Weyr, Wiener Berichte, vol. lxix. = (1), 256 THEORY OF RESIDUATION III. The partial intersection of two cubic surfaces, when the residual curve consists of a conic and a straight line lying in a different plane. Let AB he the straight line ; and let the conic be the intersection of the plane a and the quadric 8; then the equations of the sextic are {py+ qB)S = (uy + vS)0L (2), (Py+QS)S = (ii'ry + v'8)a (3), where P, p, Q, q are constants, and lo, v, u' , v' are planes. Eliminating 8 and a, we obtain (^7 + qh) {u'y + v'6) = (uy + vh) (Py + QS) (4). Equation (4) represents a cubic surface on which AB is a nodal line, and it also contains another line EF which is the residual intersection of the plane py + qB = and the quadric uy + v8 = 0. Also since AB and EF lie in the same plane, they are generators of opposite systems on the quadric. The sextic may therefore be regarded as the partial intersection of the cubics (2) and (4), which contain two straight lines lying in the same plane, one of which is a nodal line on the second cubic. IV. The partial intersection of two cubics, when the residual intersection consists of three straight lines lying in different planes. Let AB and CD be two of the lines, then the equations of the third line may be taken to be \a + B = 0, fxjB + y = 0, and the equation of the quadric having these three straight lines for generators is A-a7 = fM^S, and the equations of the two cubics which contain the sextic may be written (Xa7 - fi^B) u = (m + /3wi ) [P (ka + B) + Q (fjL^ + 7)}. . .(5), (Xa7 - fji^B) u = (av/ + ^w^) [P' (Xa + B) + Q' (fi^ + y)}.. .(6), where u, u' are arbitrary planes; Vi,w-^ ... are linear functions of (7, B\ and P, Q, P', Q' arbitrary constants. V. The partial intersection of a (juadric and a quartic surface, when the residual intersection consists of two straight lines lying in different planes. TWISTED SEXTTC CURVES 257 The equations of the surfaces containing the curve may be written ay + ^B = (I), (Pa + Q^)y + {Ra + S^)S = , (8), where P, Q, R, S are quadric surfaces. 387. In considering the possible intersections of two cubic surfaces, we have the following additional cases to consider. (i) When the two cubics osculate one another along a line AB. By virtue of § 384, this curve is the same as the partial intersection of two cubic surfaces which possess a common straight line, which is a triple line on one of them ; but since the only cubic of this species consists of three planes intersecting in the line, the sextic is an improper one consisting of three conies lying in different planes which intersect in a line. (ii) When one of the cubics has a nodal line, and the other cubic contains the line and is touched along it by one of the nodal tangent planes to the first cubic. By the corollary to | 384, this is of the same species as (i). (iii) When one of the cubics touches the other along a line, and intersects it along a third line lying in a different plane. (iv) When the two cubics intersect in two straight lines lying in different planes, one of which is a nodal line on one of the cubics. The equations of the two cubics in (iii) are a'7 + 2a^Vi + /328 + ava +^w^ =0 (9), a27 + 2a;8vi + /S-a + av/ + /3w2' = (10), whence by subtraction a (^2 — O + /S (Ws - w.') = (11), which shows that the curves (iii) and (iv) are identical. Write (11) in the form aco, + ^co,' = (12), by virtue of which (9) may be written (ary + /3vi + Va) ft)/ = (/SS + avi + Wg) &>2 (13). Let the capital letters denote what these quantities become when S = k, 7 = 1; then the sections of (12) and (13) by the plane 8 = ky are aUg + /sn^' = B. 17 I ...(14), 258 THEORY OF RESIDUATION which are the equations of two planes. The equation B = ky combined with (14) determines the six points in which the plane intersects the sextic curve ; and since only one of them lies outside AB, it follows that five of them lie on this line, which is therefore a quinquesecant. To determine these points put 7 = in the second of (14), and eliminate (a, /3) and we obtain which is a quintic equation for determining k, and shows that the five points are distinct. Also since a curve cannot in general have a quinquesecant, this species is a special kind of a more general one. By means of § 384 or directly, it can be shown that when (i) a quadric surface passes through a nodal line on a quartic surface, or (ii) a quadric and a quartic surface touch one another along a line, the residual sextic belongs to species V. CHAPTER IX SINGULAR TANGENT PLANES TO SURFACES 388. The theory of the singularities of plane curves is com- paratively easy, owing to the fact (i) that such curves possess only four simple singularities, viz. the node and the cusp which are point singularities, and the double and the stationary tangent which are line singularities ; (ii) that the two simple point singu- larities are the reciprocal polars of the two simple line singularities. But the theory of the singularities of surfaces is much more difficult, (i) because surfaces possess two simple point singularities, viz. the conic node and the binode, and six simple plane singulari- ties, the nature of which has been explained in § 11 ; (ii) because the reciprocal polar of a conic node or a binode is a compound plane singularity of a special kind, and no theory of reciprocation exists between the simple point and plane singularities of surfaces analogous to the corresponding one for plane curves. When the surface is anautotomic, the values of OTj, -sto and -575 were first obtained by Salmon* ; those of -574 and Wg by Schubert f ; but the value of -OTa appears to have been first given by myself^ in 1908. In a subsequent paper § I obtained the values of the six singular planes, when a surface possesses G conic nodes and B binodes which are isolated ; but certain portions of this investiga- tion are subject to the limitation, that the double points must not be so numerous as to cause the tangent cone from any one of them to degrade into an improper cone. These portions do not therefore apply to quartic surfaces possessing more than 11 conic nodes, since the tangent cone from a conic node being a sextic one would degrade. Cayley in his paper on reciprocal surfaces] | has attempted ■" Trans. Roy. Irish Acad. vol. xxiii. p. 461. t Math. Annalen, vol. x. p. 102 ; vol. xi. p. 348. X Quart. Jour. vol. xl. p. 210. § Ibid, vol. xLii. p. 21. II Phil. Trans, vol. clix. p. 210 ; C. M. P. vol. vi. p. 329, see p. 347. 17—2 260 SINGULAR TANGENT PLANES TO SURFACES to find the value of -33-5, which he calls j3' , for a surface possessing a nodal and a cuspidal curve of degrees h and c respectively and also C conic nodes and B binodes ; but the investigation is not very intelligible. Amongst special results, we may notice that Berzolari* found that a quartic surface having a nodal conic possesses 40 triple tangent planes, see § 274; while Pascalf states that for such a surface tn-g = 52, see § 275. The same author also states that when a quartic surface possesses 12 conic nodes, cj-g = 0, ■BTg = 32 ; and the last result agrees with that given by my own formula. In §§ 10 and 11 the various curves and developab'les connected with this branch of the subject, as well as the notation employed, have been defined and explained ; and we shall commence with a discussion of the spinodal, the flecnodal and the bitangential curves and the surfaces associated with them. I shall denote the spinodal, the flecnodal and the bitangential developables by the symbols Dg, Df, Bi,, and their edges of regression by Eg, E/, E^. The Spinodal Curve. 389. The surface a^'-^h + a"-2 [h'^Vo + S (^yS + ^7) + ^7^] + a^'-^u^ + . . . w^ = 0. . .(1 ) is one on which J. is a point on the spinodal curve, ABC is the tangent plane at A, and AB is the cuspidal tangent to the section of the surface by the plane ABC. The first step is to examine the intersection of (1) and its Hessian at A. The Hessian will be found to be of the form -8r(w-l)■j(?^-l)^'-3(7^-2)J}2S a^'^-»+... = 0...(2). Let W3 = P/33 + 3(Q7 + i^S)y82+..., then the equation of the tangent plane to the Hessian at A is 2(n-l)(Py8 + Q7 + E8)-(n-2)^^a = (3), and the tangent line AE io the spinodal curve is the intersection of (3) with the plane h, and therefore does not coincide with AB. When P = 0, the section of (2) by the plane S is the curve ra'^-2 7^ + (/3, 7)^ a»^-'7 + (/3, 7)' aJ"-' + . . . = (4), * Annali di Matematica, Serie II. vol. xiv. p. 31. t Repertorio di Moteynatiche Superiori, vol. ii. p. 424. THE SPINODAL CURVE 261 SO that the point of contact is a tacnode on the section. In this case the tangent at the tacnode is the tangent to the spinodal curve. The spinodal curve does not possess any stationary tangents, for such a tangent must have tritactic contact with (1) and also with the Hessian at the point of contact. Now the tangent cannot have tritactic contact with (1) except at the tacnodal points, where the contact is quadritactic ; but at such points it appears, from (2) and (3), that the contact with the Hessian is bitactic. Hence t = 0. We have shown in § 55 that when a straight line lies in the surface it touches but does not intersect the spinodal curve ; from which it follows that t is, in general, zero. The spinodal curve cannot have any double points ; for at such points it is necessary that the Hessian should touch the surface, which requires that P=Q = 0. The section of the surface by the tangent plane is now of the form ^^n-sy + (/3, 7) a»-^72 + (^^^ ^y ^n-4 + . . . = (5), and consequently the singularity at A on the section is the particular kind of tacnode formed by making the two tangents at a hiflecnode coincide. Now four conditions must be satisfied in order that the point of contact of the tangent plane should be a singularity of this character, which is in general impossible since the equation of a plane contains only three constants. Hence Let us now denote the degree of the original surface by N ; then the characteristics of the spinodal curve and the developable enveloped by its osculating planes are obtained from equations (10) to (15) of § 107 by writing M = 4>(N-2), S = K = T=i = (6), accordingly n = 4) I These formulae are the most important. The values of x and y can be obtained from (14) and (15) and that of ^ + ot from (13) of 262 SINGULAR TANGENT PLANES TO SURFACES §107. Further investigation is required before the values of ^ and ZT can be determined. With this exception we have obtained the characteristics of the spinodal curve and the developable enveloped by its osculating planes. The Spinodal Developable Dg and its Edge of Regression Eg. 390. This developable may be regarded indifferently as the envelope of the tangent planes to the surface at points on the spinodal curve ; or as the developable generated by the cuspidal tangents to the section of the surface by the tangent planes at these points. 391. The degree v of the spinodal developable is given by the equation v = 2N(N-2)(SN-4>) (8). Let L be any fixed line, any point on it ; then N(N — 1) (iV— 2) stationary tangents can be drawn from to the surface ; hence as moves along L these tangents will generate a scroll on which L is a multiple generator of order JSf (JSf - 1) (N — 2). Let OP be any generator of this scroll, (/, g, h, k) the coordinates of the point P where it touches the surface ; then since OP lies in the tangent plane and the polar quadric of P, it follows that if we eliminate (a, /3, 7, S) between the equations of the two planes which deter- mine L and also those of the tangent plane and polar quadric of P, we shall obtain a relation between (/ g, h, k) which is the equation of a surface X intersecting the original surface in the locus of P. Let U be the original surface ; let U (f g, h, k)= U' ; and let the equations of L be Pa + Q^ + Ry + SS=^0] pa+ q^ + ry + sB = OJ also let A' denote the operator A' = cxdldf+ I3d/dg + yd/dh -f Bd/dk (10), then the equations of the tangent plane and polar quadric at P are ^'U' = 0, A''U' = (11), and since the result of eliminating (a, ^, 7, B) between (9) and (11) furnishes an equation of degree 2(iV— l) + iV— 2 = 3iV— 4 in (/, g, h, k) this is the degree of the surface S. THE SPINODAL DEVELOPABLE 263 Now if P be one of the points where 2 cuts the spinodal curve, the generator of the spinodal developable which passes through P is also a generator of the scroll and therefore passes through the line L; hence the number of such generators is apparently equal to 4!N{N— 2) (3iV— 4) ; but since the tangent plane to thesurface at P touches the polar quadric of P (which is a cone) along a generator, this number must be halved, which gives (8), 392. The class m of the spinodal developable is given by the equation m = ^N{N-l){N-2) ..(12). Let be any arbitrary point ; then every tangent plane to the surface through 0, which touches it at a point P on the spinodal curve, is a tangent plane to the developable. Hence its class is equal to the number of points in which the first polar of inter- sects the spinodal curve. 393. Equation (12) determines the class of the edge of re- gression Es of Ds, and we must now consider this curve. If Eg had any double or stationary tangents, these would give rise to nodal and cuspidal generators on Dg, and therefore to nodes and cusps on the spinodal curve ; and since we have shown that this curve has no double points when the surface is anautotomic, it follows that T = t = 0. If, however, the surface were autotomic, t and I need not be zero. At a tacnodal point, the tacnodal taagent on the section is the tangent to the spinodal curve, and is therefore equivalent to the cuspidal tangent at two points P and P' which ultimately coincide. Hence at such a point two osculating planes to Eg coincide, and therefore the tangent plane to the surface at a tacnodal point osculates Dg along the tacnodal tangent, and is therefore a stationary plane a to Eg. Now a tacnode is a compound singu- larity which has several penultimate forms. In particular, it may be regarded as a cusp whose cuspidal tangent has quadritactic contact at the cusp, or as a flecnode whose two tangents coincide. Hence the tacnodal points are points where the cuspidal and flecnodal curves intersect. We shall hereafter prove that the flecnodal curve is the complete intersection of the surface and one of degree lli\^— 24; hence the spinodal and flecnodal curves apparently intersect in 4iV(iV— 2)(lliV— 24) points. We shall also show that these two curves touch one another, but do not 264 SINGULAR TANGENT PLANES TO SURFACES intersect ; accordingly the number of tacnodal points is half this number, and a direct proof may be given by means of the theory of united points, which has been explained in Chapter III. 394. The number of singular tangent planes whose point of contact is a tacnode on the section is ^, = 2N(N-2)(nN-24<) (13), also each of these planes is a stationary plane a- to the curve Eg, Let L be any fixed line ; through L draw a plane oc cutting the spinodal curve in a series of points P; then the tangent to the cusp at P on the section of the surface by the tangent plane at P will cut the surface in JV — 3 points Q', through L and Q draw a series of planes y; and take x and y as corresponding planes. Since there are n points P lying in the plane x, it follows that there are n{N — 3) planes y corresponding to a single plane x; hence fi = n{N-S). The spinodal developable intersects the surface in the spinodal curve three times repeated and in a residual curve of degree n', where n' + Sn = Nv, and since n' planes x correspond to each plane y, it follows that \ = n' = Nv- 3w. United planes will occur : — (i) When P is the point of contact of one of the planes ta-g. (ii) When the line PQ passes through the line L; but since each line FQ contains iV^— 3 points Q, and v is the degree of the spinodal developable, the number of united planes due to this cause is {N — 3) v. We thus obtain \^-fi = Nv + n{N-Q) = vT,^-{N-^)v. Substituting the value of v from (8) and recollecting that n = 4) \ 7ri = 4^N'(N-l){N-2) ^^^^^ o- = 2i\r(i\r-2)(lliV-24)| and by means of (4) and (5) of § 104 the following additional formulae can be obtained, viz. n= 4!N(N-2){7N-15) K = 10N'{N'-2){7N'-W) g+is= 2N'{N-2)(4!N'-1QN' + 20N'-21N' + S9) h + 8= 2N(N-2){196N'-12S2N^ + 2580N^-18Q1N+U5)j (15). The formulae (14) and (15) agree with those obtained by Salmon* by a different method with this exception. The co- efficient of the last term in the last of equations (15) is, according to Salmon, 274 instead of 270 ; and he has also assumed, without proof, that zr and 8 are zero. The Flecnodal Curve, its Developable Df and the Edge of Regression Ef of the Latter. 396. The flecnodal curve has been defined in | 10; and there are three species of singular points lying on it. In the first place the points, where the curve touches the spinodal curve, are the tacnodal points which have already been considered. In the second place the hiflecnodal points, where the planes •074 touch the surface, are nodes on the flecnodal curve, for at such points two generators of the flecnodal developable intersect on the curve. The latter cannot, however, have any cusps, for such a singularity could only occur if the point of contact of the section by the tangent plane were the particular kind of tacnode which is formed by the coincidence of the two tangents at a biflecnode ; and we have shown that such points cannot in general exist. In the third place the points, where the planes tn-g touch the surface, lie on this curve. * Geometry of Three Dimensions, p. 580. 266 SINGULAR TANGENT PLANES TO SURFACES 397. The degree of the flecnodal developable is determined by the equation v = '2N{N-S){^N-2) (16). Let ^ be a point on the flecnodal curve ; ABG the tangent plane at A ; AB, AG the tangents at A to the section by the plane ABG, of which AB is the flecnodal tangent. Then the equation of the surface TJ is + a'*-%4 + ...w,i = (17). Writing down the polar quadric and cubic oi A, it follows that both the tangents at A to the section lie in the polar quadric, and that the flecnodal tangent AB lies in the polar cubic. Let P be any point (/, g, h, k) on the flecnodal curve ; PO the flecnodal tangent to the section by the tangent plane at P. Then the equations of the tangent plane, the polar quadric and cubic of P, are A'U' = 0, A''U' = 0, A"U' = (18), where A' is given by (10); also, since (/, g, h, k) lies on the surface, U' = (19). The point P lies on the four surfaces (18) and (19), whilst (a, /8, 7, 8) are the coordinates of any point on the line PO, which is common to the three surfaces (18); if, therefore, we eliminate (/> 9> ^y ^) between (18) and (19), we shall obtain a relation between (a, /3, 7, S) which connects the coordinates of any point on the flecnodal tangent PO, and is therefore the equation of the flecnodal developable. By the usual rule, the degree of the eliminant in (a, /S, 7, S) is apparently equal to n {11 - 2) {n - 3) + %i (n - 1) (n - 3) + Sn (n - 1) (w - 2) = 6n' - 22n' 4- I8n (20), but we shall now show that this result must be reduced by 6n. Equations (18) and (19) may be regarded in another light; for if (a, /8, 7, 8) were a fixed point on the flecnodal tangent at a point P on the surface, and (/, g, h, k) a variable point, equations (18) would be the first, second, and third polars of the surface with respect to 0. Hence the result of eliminating (/, g, h, k) between (18) and (19) gives the locus of points, such as 0, whose first, second, and third polars intersect on the surface U, and the DEGREE OF FLECNODAL CURVE 267 degree of this locus is given by (20). But if we write down the first, second, and third polars of (17) with respect to A, it can easily be shown that they intersect in six coincident points at A ; hence the original surface U six times repeated forms part of the locus. Accordingly the degree of the residual surface, which is the developable in question, is Qn^ - 22^,2 + 18?? - 6w = 2n {n - 3) (3?i - 2). Changing n into N, we obtain (16). 398. The flecnodal curve is the complete intersection of the surface U and one of degree lliV— 24, and the degree of the flecnodal curve is* n = N'(llN-24>) (21). Let (a, /3, 7, B) be any point on Df. Then the result of eliminating these quantities between (18) and the equation Df («, /8, 7, 8) = gives a relation between (/, g, h, k) of degree Qv (iV - 1) + 3z/ (i\^- 2) + 2z/ (iV - 3) = 1/ {UN- 18), hence lliV— 18 is the degree of a surface which contains the flecnodal curve. But if be regarded as a fixed point on Df, and (/, gy h, k) or P a variable point, (18) may, as in the last section, be regarded as the first, second, and third polars of U with respect to ; and since these surfaces intersect U in six coincident points, which lie on the flecnodal curve, the eliminant will furnish a locus which includes Df six times repeated. Hence if F be the residual surface, we must have v{llN-l^) = vF-^Qv, giving i'^=lliV-24, so that the degree of the flecnodal curve is given by (21). 399. The class of the flecnodal developable is m = i\r(i\r-l)(lli\r-24) (22), for this is equal to the number of points in which the first polar * Otherwise thus. The point (a, /3, 7, 5) is common to the four surfaces 77 (a, /3, 7, S) = and (18) ; and if we ehminate (a, j3, 7, 5), we obtain a quantic of (/> S'l h, k) of degree lln - 18, which, when equated to zero, gives a surface which contains the flecnodal curve. But the point (a, /3, 7, 5) six times repeated is common to these four surfaces, hence U^ forms part of the locus ; accordingly the degree of the residual surface is lln -24. The form of this result shows that the locus consists of the original and residual surfaces, and the intersection of these two surfaces determines the flecnodal curve. 268 SINGULAR TANGENT PLANES TO SURFACES of U, with respect to any arbitrary point, intersects the flecnodal curve. 400. The flecnodal and spinodal curves touch one another but do not intersect. Since a tacnode may be regarded either as a particular kind of flecnode or cusp, and therefore partakes of the character of both singularities, the points of contact of the tangent planes -as-g must be the points where the spinodal and flecnodal curves intersect one another. The total number of these points is but since the number of planes ■575 is half this number, the two curves must touch one another at the points of contact of tn-g. 401. The 27 lines lying in an anautotomic cubic surface constitute the flecnodal curve ; also any line lying in a surface of higher degree forms part of this curve, and the theorem of § 55 is a particular case of the preceding one. If the flecnodal curve consists entirely of straight lines lying in the surface, their number is iV(lliV"— 24), hence : — A surface of the Nth degree cannot possess more than iV^(lli\^— 24) straight lines lying in it. 402. Before explaining Schubert's method for finding the number of planes -33-4 and ■sTg, some preliminary theorems will be necessary. The flecnodal developable intersects the surface in a residual curve of degree nf=2N(N-4<)(SN' + N-12) (23). Let n, Vf denote the degrees of the flecnodal curve and developable respectively ; then since the developable intersects the surface in the flecnodal curve four times repeated, the degree nf of the residual curve is given by the equation Nvf=4,n + nf (24). Substituting the values of ly and n from (16) and (21) we obtain (23). 403. If P be any point on the flecnodal curve, the ordinary tangent at P to the section of the surface by the tangent plane at P, generates a developable whose degree vq is v, = N{nN^-d^N+M) .: (25). THE TANGENT PLANE t^, 269 Let P be any point on the flecnodal curve, then the flecnodal and ordinary tangents at P will generate two developable surfaces Vf and Vi^ ; but if P be one of the points where the flecnodal curve intersects the surface S, which has been discussed in § 391, one of these two tangents must intersect the fixed line L. Accordingly the degree of the compound surface generated by both tangents is vf+Vo = N{llN-^^) (SN - 4). Substituting the value of v/ from (16) we obtain (25). 404. The surface vq intersects the original surface in a residual curve of degree Uq, where no = Nvo-nn (26). For every generator of Vo intersects the surface in the flecnodal curve three times repeated and in a residual curve Wq. 405. The number of singular tangent planes, whose point of contact is a biflecnode on the section, is ^, = 5N{7N'^-28li+S0) (27). A plane x through a fixed line L intersects the flecnodal curve in n points, where n is given by (21). Let P be one of them, then the ordinary tangent to the surface at P intersects it in N—S points Q, all of which lie on the curve oi^. Let the planes through L and the points Q be the planes y, and take cc and y as corre- sponding planes. To every point P correspond iV — 3 planes y ; and since there are n points P, there are (iV— 3)n planes y corresponding to a single plane x ; hence fx = (N-S)n. A plane y intersects the curve Uq in Wq points, to each of which corresponds a plane x ; hence \ = nQ. United planes will occur : — (i) When one of the points Q coincides with P, in which case P is a point of contact of a tangent plane ■3x4. But since both the tangents at P are flecnodal ones, and Q may be supposed to coincide with either of them, this plane must be counted twice ; hence the number of united planes due to this cause is 2'sr4. (ii) When P is a tacnodal point, one of the points Q will 270 SINGULAR TANGENT PLANES TO SURFACES coincide with P ; hence the number of united planes due to this cause is -zn-g. (iii) Let P be a point where the ordinary tangent intersects the line L ; then since iV— 8 points Q lie on this tangent, the number of united planes due to this cause is {N—^)vq. We thus obtain \-\-ti = n, + {N-'^)n = 2-374 + ^3-5 + (iV- 3) v,. Substituting the values of 72o) v^, n and 075 from (27), (26), (21) and (13) we obtain (27). 406. The number of tangent planes, whose point of contact is a hyperfiecnode on the surface, one of whose tangents has ordinary contact and the other quadritactic contact with their respective branches, is -5r6=5iV(i\^-4)(7iV-12) (28). The planes oc are the same as before ; but the points Q are those where the flecnodal tangent at P intersects the surface ; hence fi = {N-4<)n. A plane y intersects the curve nf in ny points, to each of which corresponds one plane oc ; hence X = nf. United planes will occur : — (i) When a point Q coincides with P, in which case P is the point of contact of a plane OTr ; hence the number of united planes due to this cause is ■arg. (ii) Let P be a point where the flecnodal tangent intersects the line L; then since iV— 4 points Q lie on this tangent, the number of united planes due to this cause is (N — 4) Vf. We thus obtain X + /i = w/ + (i\r - 4) w = OTfi + (iV - 4) Vf. Substituting the values of n./ and Vf from (24) and (23) we obtain (28). 407. The Flecnodal Curve. The characteristics of this curve and the developable enveloped by its osculating planes can now be partially found by means of equations (10) to (15) of § 107 ; for we have shown in | 398 that ilf=lli\'^-24 (29), THE BITANGENTIAL CURVE 271 ■Whence, by (9) of § 107, 2/i = i\^(i\^-l)(lliV'-24)(lli^-25) (30). The points of contact of the planes W4 are nodes on the curve, whence, by (27), we obtain B = 5N(7N'-28N+'S0) (31). Substituting these values and recollecting that k = 0, we obtain V = 2N (N - S) (SIN - 54) TO = 3i\^(62i\^^-305i\r+348)-t I (32), (T = 4>N (93i\^2 _ 4(j3jyr + 534) _ 2t and the value of ?/ + r can be found from (15) of § 107. The first of (32) gives the degree of the developable enveloped by the osculating planes to the flecnodal curve ; but whether or not the curve possesses any points of inflexion cannot be ascertained without further investigation. It appears to me possible that the points of contact of tn-g might be points of this character. 408. The Flecnodal Developable and its Edge of Regression. Our knowledge of this surface and curve is confined to the equations z/=2J\^(i\r-3)(3iV-2) \ 7n = N {N -1) {l\N -^^Yr (33), ^ = 5N {IN' - 28N + 30) j r = t=0 (34). The third of (33) arises from the fact that the planes ■374 are double tangent planes to Df, and therefore doubly osculating planes to Ef. The planes tn-g also, in all probability, give rise to some singularity. The Bitangential Curve, its Developable D^ and the Edge of Regression Ej^ of the Latter. 409. The class of the bitangential developable is m = ^N{N-l)(N-2){N'-N' + N-12) (35). Let be the vertex of the tangent cone to the surface from an arbitrary point ; then every double tangent plane to the cone is a tangent plane to D^ ; hence m is equal to the number of double tangent planes to the cone. Let u, fi be the degree and class of 272 SINGULAR TANGENT PLANES TO SURFACES the cone ; B, k the number of its nodal and cuspidal generators ; then, by Chapter I, v = n{n—l), fi = n (n — ly, 8 = ^n{n-l){n-2){n-S), K = n(n-l){n-2), whence, by Pliicker's equations, we obtain 2T = n (n - 1) (n - 2) (n^ -ii' + n- 12). Changing r into m and n into N, we obtain (35). 410. The degree of the hitangential curve is n = N(N-2){N'-N'' + N-12) (36). Let T be the degree of the hitangential surface, that is the surface which intersects the original one in the hitangential curve ; let OPQ be a double tangent plane to the tangent cone from 0, which touches the cone along the generators OP, OQ ; and let P and Q be the points where these generators touch the surface U. Then the number of points such as P and Q is obviously equal to 2m; but these points are the intersections of the hitangential surface, the original surface and its first polar with respect to ; hence their number is equal to TN{N— 1). Accordingly TN(]S'-l) = 2m. " Substituting the value of m from (35), we obtain T = (N-2)(N'-N' + N-12) (37). Equation (37) gives the degree of the hitangential surface, and the degree of the hitangential curve is this quantity multi- plied by N. 411. The spinodal and hitangential curves touch one another at the tacnodal points. They intersect one another at the points which are the cuspidal 'points on the tangent planes tn-j; and the number of such planes is ^, = 4>N(N'- 2) (N-S) {N' + SN-16) (38), also the planes ■CTi are stationary planes to the edge of regression of the hitangential developahle. Let P and Q be the points of contact of any double tangent plane to the surface ; then P and Q are nodes on the section by the plane. But a tacnode may be formed by the union of two nodes, hence if P and Q coincide, P becomes a tacnodal point on the surface, and the tacnodal tangent PQ becomes a tangent to THE TANGENT PLANES Z!T^ AND ttt. 273 the bitangential curve. Hence the bitangential curve touches the spinodal and also the flecnodal curve at the tacnodal points. The tangent plane ■bti touches the surface at a point p, which is a cusp on the section, and at another point q which is a node. Hence p must be a point where the spinodal and bitangential curves intersect one another. Accordingly the number of such points plus twice the number of tacnodal points is equal to the number of points in which the spinodal and bitangential curves intersect; whence 'ST, + 4>N{N' -2) (UN -U) = 4^N(]Sr-2y(N'-N' + N -12), giving ^, = 4>JV{N-2)(N-S){N^ + SN-16). To prove the last part of the theorem, let ABG be the plane -571 ; B the node, A the cusp, and AG the cuspidal tangent in the section. Then if we write down the equation of the surface and its first polar with respect to any point T on AB, and then put 8 = 0, we shall obtain exactly the same equations as if we had first put 8 = 0. Hence these equations represent the section by the plane 8 of the surface and of its first polar with respect to T ; and we know from the theory of plane curves that these two sections have tritactic contact with one another at A, and that AG is the common tangent. Hence AG is the tangent to the bitangential curve at A, and the generator AB of Dj is equivalent to three coincident generators through three coincident points at A. From this it follows that the plane 8 or ■m-^ osculates D^ along AB, and is therefore a stationary plane to Ef). We thus obtain the equation (T = ^, (39). We have also proved that : — The tangent to the bitangential curve at a point, where it intersects (hut does not touch) the spinodal curve, is the tangent at the cusp on the section of the surface hy the tangent plane ■OTi. 412. The points, where the bitangential and flecnodal curves intersect one another, are the flecnodal points on the tangent planes •zB-g ; and the number of such planes is ^, = N{N-2) (UN -24'){N'-N' + N-16).. .(40). The tangent plane •ur^ touches the surface at a point P, which is a flecnode on the section, and at another point Q, which is a B. 18 ■ 274 SINGULAR TANGENT PLANES TO SURFACES node. Hence P is a point where the bitangential and flecnodal curves intersect one another. Accordingly ^. + 4i\^(i^-2)(lli^-24) = N (N -2)(11N -24^){N'' - N' + N -12), giving ^, = N(N-2){nN-24<){N'-N' + N-16). Reciprocal Surfaces. 413. Let 8 be an anautotomic surface of degree n, S' the reciprocal surface; and let the unaccented and accented letters refer to the original and the reciprocal surface respectively. Let T be any plane section of >S' ; then, since the characteristics of an anautotomic plane curve are m = n{n-l), t = |n (n - 2) (n^ - 9)| . t = nn(n-2), 8 = 0, k = J the reciprocal of a plane section of ^ is a tangent cone to S', whose characteristics are n =n{n—l), m' = n, S' = ^n {n — 2) (n^ — 9)| , K' = Sn{n-2), t' = 0, l=0 ]'" (i) Let the plane T have ordinary contact with >S' at a point 0. Then is a node on T, and therefore S = 1, t = 1 ; also the vertex 0' of the cone lies. on 8', and the double tangent plane to the cone is the tangent plane to 8' at 0'. The two generators along which this plane touches the cone are the nodal tangents at 0' to the section of 8' by the tangent plane, and they are the reciprocals of the nodal tangents to T at 0. (ii) Let be a cusp on T. Then k = 1, and if = 1; hence the tangent plane at 0' to 8' osculates the cone along a generator. Through 0' draw an arbitrary plane P', then the reciprocal of P' is a point P lying in the plane T ; and the reciprocal of the section of 8' by P' is the tangent cone to 8 from P. Now the plane T can easily be shown to osculate this cone along a generator PO* ; * For the purpose of proving this result, it is sufficient to employ the cubic surface a^8 + a (d^vo +Svi+ py^) + !/3 = 0. The tangent plane 5 touches the surface at A, which is a cusp on the section ; also C is any point in this plane. Writing the cubic in the binary form (7, 1)''=0, and equating its discriminant to zero, we obtain the equation of the tangent cone from C, which shows that the plane ABC osculates the cone along the generator ^C. I (43). RECIPROCAL SURFACES 275 hence 0' is a cusp on the section oi 8' hj P\ and consequently the locus of 0' is a cuspidal curve on S', which is the reciprocal of the spinodal developable of 8. The characteristics of the latter are given by equations (14) and (15); hence, reciprocating, we obtain the following formulae for the cuspidal curve on 8', viz. v = 2N(N-2){2N-4^) oi = ^N(N-l)(N-2) K = 2N (N - 2) (UN - 24^) m = ^N{N-2){1N-\h) a = 10N {N -2){1N -IQ) The remaining characteristics can be obtained from the last three of (15) by writing y, h, B, g and ot for x, g, w, h and 8 respectively. These formulae show that the tacnodal points on 8 correspond to cusps on the cuspidal curve on 8'. The reciprocal of the spinodal curve is the developable en- veloped by the tangent planes to 8' at points on the cuspidal curve. The characteristics of this developable and of its edge of regression are obtained by reciprocating (7) and the last four of (6). (iii) Let T be a double tangent plane, and let P and Q be its points of contact. Then 3 = 2, and t' = 2 ; hence the cone has a pair of double tangent planes, both of which are tangent planes to 8' at 0'. Accordingly the locus of 0' is a nodal curve on 8', which is the reciprocal of the bitangential developable. The characteristics of the latter have only been partially obtained ; but by reciprocating (35) and (38), and recollecting that ■OTj is a stationary plane to Ej,, and therefore gives rise to a cusp on the nodal curve, we obtain the following formulae for the nodal curve on^'. n = ^N{N-l)(N-2)(N''-N'' + N-12)) The reciprocal of the bitangential curve on 8 is the developable enveloped by the tangent planes to 8' at points on the nodal curve. (iv) Let be a flecnode on T. Then S = 1, t = 1 ; so that t'' = 1, k=1. Hence the cone has a cuspidal generator, whose 18—2 276 SINGULAR TANGENT PLANES TO SURFACES cuspidal tangent plane touches the cone along another generator, and is therefore a double tangent plane. This plane is the tangent plane to S' at 0', and the locus of 0' is a curve on S', which is the reciprocal of the flecnodal developable. Its character may be investigated by means of the quartic surface + 3a (S%o + ^^w, + Sw, + y F2) + W4 = 0. . .(45) or a^S + Sa^u^ + Sau^ + W4 = 0, in which the plane S touches the surface at a point A, which is a flecnode on the section, and AB is the flecnodal tangent. The tangent cone at A is from which it follows that the plane S is a double tangent plane, which has ordinary contact along the generator AG, but AB is a cuspidal generator whose cuspidal tangent plane is S. This shows that 0' is a point on the flecnodal curve on 8', accordingly : — The reciprocals of the flecnodal curve and developable on 8 are the flecnodal developable and curve on 8'. Reciprocating (33), the degrees of the flecnodal curve and developable on 8' are n = N(N-l){UN-2^)^ v = 2N{N-S){d]S[-2) J ^ ^" To avoid circumlocution, I shall denote the degrees of the nodal, cuspidal, and flecnodal curves on 8' by the letters b, c, and/j and shall frequently refer to them as the curves b, c, and/"; whilst the degree of the bitangential curve on 8 will be denoted by p. By (44), (43), (46), and (36) their values are b = ^N{N-l)(N-2)(N'-N"^ + N -12\ c = 4>N{N-l){N-2) I ^4^^_ /= N{N-l){llN-24>) [ p= N(N-2){N'-N^ + N-12) f Moreover, it is possible for the nodal curve, considered as a curved line drawn on the surface, to possess nodes, cusps, and other singularities ; and these must be carefully distinguished from singular points, such as pinch points, which are singular points on the surface, but not necessarily such on the curved line, which constitutes the nodal curve. RECIPROCAL SURFACES 277 (v) Let be a point where the bitangential and spinodal curves intersect ; then the plane T has ordinary contact with 8 at some point Q on the former curve and is the double tangent plane tsT^, one of whose points of contact is a cusp on the section, and its cuspidal tangent is some line OP, whilst Q is a node. Hence S = 1, /c = 1 ; and therefore r = 1, l =1. Accordingly the cone has a double and a stationary tangent plane, both of which touch S' at 0'. The latter plane is the cuspidal tangent plane at 0' to S' along the curve c, and is one of the nodal tangent planes to the curve h ; whilst the former plane is the other nodal tangent plane at 0' to the curve h. From this it follows that 0' is a cubic node of the fifth species on S'. We have also shown in (iii) that 0' is a cusp on the curve h ; but since the generators OP and OQ of the spinodal and bitangen- tial developables to >Si at do not, in general, coincide, the curves h and c on 8' intersect, but do not touch at 0'. Furthermore, if P' , Q' be two points on the curve b near 0', the curve c cuts the plane O'P'Q' at a finite angle ; but if O'R is the cuspidal tangent at 0' to the curve h, the three tangent planes to the surface at 0' all pass through O'R'. (vi) Let be a tacnode ; then is a point where the spinodal and bitangential curves touch, and T is the singular tangent plane •sTg. Also S = 2, T = 2 ; so that 8' = 2, t' = 2 ; accordingly the cone has a tacnodal tangent plane, which is the tangent plane to 8' at 0'. In this case the curves h and c touch one another at 0', and 0' is a pinch, point on the former. Moreover, from (ii), 0' is a cusp on the curve c, and the two coincident nodal tangent planes to h coincide with the cuspidal tangent plane to c at 0'. These three coincident planes pass through the cuspidal tangent to c at 0' ; and 0' is a cubic node of the sixth species on 8'. Any plane section through a cubic node of the sixth species has a triple point of the third kind thereat ; and we can verify this by the method explained in (i) by means of the quartic surface ,y4 + 4y (^avo + v^) + 672 (aX + olw^ + ^2) + ^7 {o^h Fo + aF^ + F3) + a='SFo + a^STf"i + aSF2+Tr4 = (48), where the suffixed letters denote binary quantics of (/3, S). The plane ABC is the tangent plane at A ; also this point is a tacnode on the section, and AB is the tacnodal tangent. The equation of 278 SINGULAR TANGENT PLANES TO SURFACES the tangent cone from C, which is any arbitrary point on the section, is obtained in the usual manner by equating to zero the discriminant of (7, iy = 0, viz. 7^ = 27/^ from which it will be found that the plane B has quadritactic contact with the cone along AG. This shows that A is a, point of undulation on the section of the cone by the plane ABD, and that AB is the tangent at this point ; and since the reciprocal of the tangent at a point of undulation on a plane curve is a triple point of the third kind, 0' is such a point on the plane section of 8'. (vii) Let be a point where the bitangential and flecnodal curves intersect ; then the plane T has ordinary contact with S at some other point Q on the bitangential curve, and T is the double tangent plane -sr^, one of whose points of contact is a flecnode on T. Hence S = 2, t = 1 ; and therefore t' = 2, k =1. Accordingly the cone has one ordinary double tangent plane corresponding to Q, and a singular tangent plane corresponding to 0, which has ordinary contact with the cone along one generator and is the cuspidal tangent plane to the cone along another generator. These two planes are the nodal tangent planes at 0', but the one corresponding to touches the flecnodal curve and the latter intersects the nodal curve at 0'. The value of OTa is given by (40). (viii) Let T be a triple tangent plane to S ; and let P, Q, R be its points of contact. Then these points are nodes on the section, and are also points on the bitangential curve. Hence S = S and t = 3. The tangent cone from 0' has therefore three double tangent planes which are tangent planes to S', at 0' ; hence 0' is a cubic node of the third kind on S', and a triple point of the first kind on the nodal curve. The number of these points will be considered later on. We have now completed the discussion of the spinodal and bitangential curves, but the flecnodal curve remains to be con- sidered. (ix) Let 0' be a biflecnode on T. Then S = l, l = 2; so that t' = 1, K =2; and by employing a similar method to that of (iii) it can be shown that the cone possesses two cuspidal generators having a common cuspidal tangent plane, and that 0' is a biflecnode on the section of the tangent plane at 0' to 8'. (x) Let be a hyperflecnode on T, one of whose tangents has quadritactic contact with its own branch, and consequently RECIPROCAL SURFACES 279 quinquetactic contact with S at 0. Then B = l, t=1, t=2; hence t =1, 8' = 1, /c' = 2 ; and the tangent plane at 0' touches the cone along two generators, one of which is a triple generator of the third kind, whilst the contact along the other is ordinary bitactic contact. To ascertain the character of the singularity at 0', let us consider the quintic surface a^S + 4a3 {8% + Svi +pl3y) + Ga^ (8u^ + ryt^) + 4a (Sw3 + 7^3) + u, = 0, where the us are ternary quantics of (^, 7, B), and the other letters are binary quantics of (/3, S). The section of this surface by the tangent plane S is the singularity in question, AB being the tangent which has quinquetactic contact with the surface ; and if the equation to the tangent cone from A be written down, it will be found that ABC is a double tangent plane to the cone along the generators AB and AG, and that AB is a triple generator of the third kind, whilst the contact is bitactic along AG. This shows that the singularity at 0' on the reciprocal surface is of the same character as that on the original one. 414. Equations (46) and (47) furnish a verification of Cayley's theorem of § 59 ; for the degree n of the flecnodal curve on the reciprocal surface is given by the equation n = M{llM-24!)-22b-27c, where M=N{N—iy. Substituting the values of b and c from (47), it will be found that this equation reduces to the first of (46). The corresponding equation, which gives the degree of the spinodal curve on S', is by § 58 7i = 4if(ilf-2)-86-llc (48 a). Now the spinodal curve on 8' gives rise to a spinodal developable, the reciprocal polar of which is a cuspidal curve on S. But since 8 is anautotomic, it possesses no cuspidal curve and therefore 8' possesses no proper spinodal curve, and the degree of the latter is therefore zero ; hence the curve of intersection of 8' and its Hessian must consist of the nodal and cuspidal curves on 8' repeated a certain number of times. And if the values of M, b and c be substituted in (48 a), it will be found that ?i = as ought to be the case. 280 SINGULAR TANGENT PLANES TO SURFACES I have not succeeded in ascertaining the reduction in the degree of the bitangential curve which is produced by a nodal and a cuspidal curve ; but if the reduction is denoted by xh + yc, the method of the preceding paragraph indicates that x and y are functions of the degree N of the surface. This is confirmed by the fact that a double point on the original surface gives rise to a multiple point of order N{N—Vf — Q on the bitangential surface*; and we should therefore anticipate that a nodal or a cuspidal curve on the original surface gives rise to a multiple curve on the bitangential surface, whose multiplicity is a function of the degree of the original surface. 415. To find the number of triple tangent planes to an anautotomic surface. We shall prove the formula 6(7i'-2) = p + 2t!rg + 8c7i + 3OT3 (49), where n' = N'{N—Vf is the degree of the reciprocal surface; ■573 is the number of triple tangent planes to 8, and p is the degree of the bitangential curve. The values of h and p are given by the first and last of (47). Let A be any point in space ; let a surface of degree n possess a nodal curve of degree h, and a cuspidal one of degree c ; also let a be the number of ordinary tangents which can be drawn from A to any plane section of >Si' through A. Then, by Pliicker's equations, a = n'{n' -l)-^h-2,c (50). The complete tangent cone from A to ;S" is of degree n{n' — V)\ and (50) shows that it consists of the cone twice repeated, which stands on the nodal curve h, the cone three times repeated, which stands on the cuspidal curve c, and a proper cone whose degree a is given by (50). Equation (49) is proved by examining the character of the points of intersection of the second polar of A with the nodal curve h. These points are ordinary and singular. At every ordinary point B, in which the curve of contact of the cone a intersects the curve h, one of the nodal tangent planes must pass through A, and we shall first show that these points lie on the second polar of A. * "Singular tangent planes to aiitotomic surfaces," Quart. Jour. vol. xui. p. 37. NUMBER OF TRIPLE TANGENT PLANES 281 Consider the surface* a" Wo + OL^'^U^ + . .. a%„_2 + CL£lslln-s-\ + ^s"Un-'ZS = 0- • .(51), where fls = yS*-lWl + )S«~■■^^f;2+ ... Wj Equation (51) represents a surface having a plane nodal curve (a, Hg), which passes through B. The nodal tangent planes at B are obtained by equating the coefficient of /3"~^ to zero, and are of the form a?Va + FottWi + WqW-^ = 0, and if one of them passes through A, Vo = 0. The second polar of A is obtained by differentiating (51) twice with respect to a, and Vq = () is the condition that it should pass through B. Let us now reciprocate this result. The point A becomes an arbitrary plane P ; the tangent cone a becomes the section of S by this plane ; the points, where the curve of contact of a intersects the nodal curve h, become the tangent planes at the points where the section of >S by P intersects the bitangential curve, and the number of these points is equal to p. It follows from (vi) that the points on 8' corresponding to tiTg are cubic nodes of the sixth kind, and such points are ordinary points on the second polarf, and the latter has ordinary contact with the surface at such points. Also the points in question are ordinary points on h, hence the second polar and this curve have bitactic contact with one another at these points. Accordingly the number of points of intersection arising from this cause is 2-575. It follows from (v) that points on /S" corresponding to Wi are cubic nodes of the fifth species on the surface ; and such points are ordinary points on the second polar, but the latter does not touch the surface. The tangent plane, however, passes through * Although the method of proof only applies to surfaces having a plane nodal curve, there can be no doubt that the theorem is true when the nodal curve is twisted. t The equations of a surface having a cubic node of the sixth species at A, and of its second polar with respect to D are and 6a'»-3 S + a™-*M4" + . . . «„"= 0, where u^" — dujdd. 282 SINGULAR TANGENT PLANES TO SURFACES the line of intersection of the tangent planes at the cubic node* on S'; and this line is the line O'R' considered at the end of (v), which is the cuspidal tangent to the curve b at 0\ Hence the second polar intersects the nodal curve b at 0' in three coincident points ; accordingly the number of points of intersection arising from this cause is Sot^. Since every multiple point of order A; on a surface gives rise to a multiple point of order k — 2 on the second polar, it follows that the second polar passes through every cubic node on the surface. Now we have shown in (viii) that every triple tangent plane gives rise to a cubic node of the third kind on S', and to a triple point of the first kind on the curve b. Accordingly the number of points of intersection arising from this cause is S^s. We have therefore proved the formula (49), and we have to substitute the values of b, p, OTg and -OTi from (47), (13) and (38); also n =N{N —Vf\ we thus obtain org = ii\r (i\r _ 2) {N-' - 4i\^« + IN' - 45iV^ + lUN' - llli\^2 4. 548i\r- 960). ..(53), which determines the number of triple tangent planes. 416. Ilie degree of the bitangential developable is v = N(N-2){N-S){N' + 2N-4!) (54). By (13) and the fourth of (47) equation (54) is equivalent to ^ = P-¥^5 (55), and we shall prove the last equation by the Theory of Corre- spondence. Through any fixed line L draw a plane a) cutting the bi- tangential curve in p points P ; and let the generator of the bitangential developable through P intersect the curve in Q; through L and Q draw a plane y, and take a; and y as corresponding planes. Then to every point P one point Q corresponds and vice versa ; hence \ = /ji = p. * This may be proved by considering such a surface as a>''-3(py + qS) 5- + a"'-itti+...u^-0, for its second polar with respect to D is 2a''-3 {py + Sqd) + a"-4 H4" + . . . M„" = 0. THE BITANGENTIAL DEVELOPABLE 283 United planes will occur : — (i) When P and Q coincide, in which case P will be the point of contact of a tangent plane Wg; hence the number of united planes due to this cause is OTg. (ii) When the line PQ intersects L. There are obviously v of such lines, but since the plane LPQ may be regarded as a plane x or y, this plane is equivalent to two united planes ; hence the number due to this cause is 2z/. We thus obtain A, + yu- = 2p = tsTg + 2z/, which is the required result. 417. We have therefore proved the following formuljB for the bitangential developable and its edge of regression, viz. a = 4'N{N-2)(N-S)(Ii-' + SN-16) - (.56), v= N(N-2)(N-S){N' + 2N'-4^) aud from (4) and (5) of § 104 we easily obtain n = ^N{N- 2) (5N' - lli\^^ + 12i\^2 - 221i\^+ 420) | which determine the degree of the edge of regression, and also the number of its cusps. The arguments that we have already used show that r and I are zero ; also, since the curve b does not possess any isolated nodes, there are no isolated planes ■sr, for these are included in the triple tangent planes tn-g, each of which osculates Ej, at three distinct points. The remaining quantities x, y, and g can be obtained from (4) and (5) of § 104. 418. To find the number k' of apparent double planes of the bitangential developable. The value of k' is equal to the number of apparent nodes of the nodal curve on the reciprocal surface. Now we have already shown that the tangent planes Wi and OTs to S respectively give rise to cusps and triple points of the first kind on the nodal curve b on S' . Also every triple point is equivalent to three actual nodes, but the curve has no other actual nodes except 284 SINGULAR TANGENT PLANES TO SURFACES those included in the triple point ; we must therefore write in (5) of § 104 and we obtain j; = h(h-l)- 2k' - 61:73 - 3t3-i, in which all the quantities except k' are known. We thus obtain - 288N' + 547i\^^ - 1058N' + 1068iV^- - 1214i\^ + 1464). . .(58). 419. The following equations give the numbers of the six singular tangent planes to an anautotomic surface of degree N; and the number attached to each equation indicates where it is to be found in the text : ■UT, = 4!N(N-2)(N-S)(N' + SN-16) (38), ^, = N{N-2){nN-24!)(N'-m + N-l(5) (40), ^3 =i]\[(]}{- 2) {N^ - 4i\^« + 7N' - 4^bN' + 114iV'«_ llliyr2 + 548i\r_ 960)...(53), ^, = bF{7N''-28N + S0) (27) ^5 = 2i\^(iV^-2)(lli\r-24) (13), ^,= 5JSf(N-4>)(1N-12) (28). The preceding analysis gives a fairly complete investigation of the six curves and developables mentioned in §§ 10 and 11, with the exception of the developable and curve Df and Ef, with respect to which further investigation is required to complete the theory. Autotomic Surfaces. 420. I shall not give any detailed account of the theory of singular tangent planes to autotomic surfaces, which possess G conic and B binodes, since the investigation is lengthy, and for the reasons stated in my paper* the results must be regarded as provisional until verified by some independent method, such as the Theory of Correspondence. Let J. be a conic node ; draw the tangent cone from A, and let the curve of contact cut the spinodal curve at P. Then since the tangent plane along the generator AP intersects the surface * f 'Singular tangent planes to autotomic surfaces," Quart. Jour, vol. xlii. p. 21. AUTOTOMIC SURFACES 285 in a curve which has a node at A and a cusp at P, the plane is a singular tangent plane of the species OTj ; but it is an im'proper plane, because the contact at A is not ordinary contact, but is of a special character due to the fact that ^ is a conic node. The true tangent planes tiJi are those which touch the surface at two points P and Q, one of which is an ordinary point of intersection of the spinodal and bitangential curves, whilst the other is an ordinary point on the latter curve. In like manner, when a surface possesses G conic nodes, the improper triple tangent planes are (i) every double tangent plane to the tangent cone from a conic node, (ii) every tangent plane to the surface through a pair of conic nodes. Similar observations apply to surfaces which possess binodes as well as conic nodes ; from which it follows that every double point on a surface must produce a diminution in the number of singular tangent planes to the surface, similar to that produced by a double point on a plane curve in the number of double and stationary tangents. Accordingly a set of formulae exists for surfaces similar to Pliicker's equations for plane curves. 421. One of Pliicker's equations for a plane curve is 3w (w - 2) = t + 68 + 8/c, in which the left hand side is equal to the number of stationary tangents possessed by an anautotomic plane curve, whilst the right hand side shows that, when the curve is autotomic, each nodal tangent is equivalent to three and each cuspidal tangent to eight stationary tangents. And by considering the surface formed, by the revolution about the axis of x, of a plane curve symmetrical about this axis which has a node upon the latter, it follows that the tangent cone from the node three times repeated forms part of the spinodal developable. Hence every conic node reduces the degree of the spinodal developable by 6 ; and in a similar manner it can be shown that a binode reduces it by 8 ; accordingly the degree v of the true spinodal developable is 1/ = 2i\r(ZV- 2)(3i\r-4)- 6C- 85. 422. In the paper referred to I have worked out the degrees and classes of most of these developables ; and the method employed in calculating the singular tangent planes is to find the number of improper tangent planes of each species, and to subtract their number from the value of -sr for an anautotomic 286 SINGULAR TANGENT PLANES TO SURFACES surface, which is denoted by vr'. The value of OTb is unaltered b}' ordinary double points, and the change produced when the singularity is a compound one has not been considered. The final results are as follows : — ^1 = <- 2 {iV(iV- 1) (7iV^- 11) - 6C- 54] C - 4 [N{N- 1) (5i\r- 8) - 65 - 36} B + 325(7, t!72 = < - 2 {i\^ (i^ - 1 ) (1 7i\^ - 30) - 1 2(7 - 84} C - 3 {iy^(iV- l)(17i^- 30) - 185- 96} 5 + 6650, t:73 = ^3'- 2(7^1 - 35^2 - 125- 2 (ilf - 8) (7(C- 1) -f(M-6)5(5-l)-6(if-7)5C, ti3-4 = <-30C-455, t:7g = <-240-365. The value of M is ti is the number of double tangent planes to the tangent cone from a conic node, and t^ the number when the vertex is a binode. Their values are 2^1 = (ilf- 9)^ - iV(i\^- 1) (3i\^ - 14) - 35 - 1, 2^3 = (ilf-8)2-i\^(iY-l)(3iy^-14)-85-10. APPENDIX I. On Plane Trinodal Quartics. Let the tangents at the node of a uninodal quartic cut the curve at B and C ; and let the line BG cut the curve in Q, Q'. Then these points, which are called the Q points, possess various important properties which have been discussed by Roberts*. He employs the parametric method, but all his results can be obtained much more simply by the ordinary methods of trilinear coordinates. A trinodal quartic possesses three pairs of Q points, of which one pair corresponds to each node ; hence Roberts' results are capable of extension to these curves. The three conies mentioned in § 194 of my treatise on Cubic and Quartic Curves pass through two points, which I call the 8 points ; and the line 8S' intersects the quartic in two other points, which I call the T points; and both pairs of points possess various important properties. Let the equation of the quartic be /Sy + 7V + a^yS^ + 2a/37 {la + to/3 + ^7) = (1) ; also let a = Ift'y + TO7a + na^, T = /37/Z + ^ajm + a^jn, u = ha + A^a/S + hy, ki = 7n/n + njm — 21, &c., then (1) can be written in the form