Conics and perspectivities

Transcription

1 10 Conics and perspectivities Some people have got a mental horizon of radius zero and call it their point of view. Attributed to David Hilbert In the last chapter we treated conics more or less as isolated objects. We defined points on them and lines tangent to them. Now we want to investigate various geometric and algebraic properties of conics. In particular, we will see how we can treat conics on the level of bracket algebra Conic through five points We start by calculating a conic through a given set of points. For this consider the quadratic equation that defines a conic. a x 2 + b y 2 + c z 2 + d xy + e xz + f yz =0. This equation has six parameters a,..., f 1. Multiplying all of them simultaneously by the same non-zero scalar leads to the same conic. Thus the parameter vector (a,..., f) behaves like a vector of homogeneous coordinates. Counting degrees of freedom shows that in general it will take five points to uniquely determine a conic. To find the parameters for a conic through five points p i = (x i,y i,z i ); i =1,..., 5 we simply have to solve the following linear system of equations: 1 Compared to Section 9.1 we have relabeled the parameters and put the factor of 2 of the mixed terms into the parameters

2 Conics and perspectivities x 2 1 y1 2 z 2 a 1 x 1 y 1 x 1 z 1 y 1 z 1 x 2 2 y2 2 z2 2 x 2y 2 x 2 z 2 y 2 z 2 b 0 x 2 3 y2 3 z2 3 x 3y 3 x 3 z 3 y 3 z 3 x 2 4 y4 2 z4 2 x 4 y 4 x 4 z 4 y 4 z 4 c 0 d = 0 x 2 5 y2 5 z2 5 x e 0 5y 5 x 5 z 5 y 5 z 5 0 f If this system has a full rank of 5 then there is an (up to scalar multiple) unique solution (a,..., f) that defines the corresponding conic. If more than 3 points are simultaneously collinear of if two points coincide the rank of the system may be lower than 5. This corresponds to the situation that there are more than one conic passing through the given set of points. This method of determining the parameter vector (a,..., f) is mathematically elegant however it is computationally expensive. We first have to calculate the squared parameters and then have to solve a 5 times 6 system of equalities. There is also another way to calculate such a conic more or less directly. This way will also give us additional structural insight into the geometry and underlying algebra of a conic. In preparation we have to understand how to calculate a degenerate conic that consists of two lines with homogeneous coordinates g and h. A conic must be represented by a quadratic form p T AP p =0 that vanishes if p is on either of the lines. The (non-symmetrized) matrix of such a quadratic form is simply given by A = gh T. This can be easily seen since the quadratic form p T Ap = p T (gh T )p =(p T g)(h T p)= p, g p, h vanish if one of the two scalar products on right side vanish. This in turn corresponds geometrically to the situation in which p is on g or on h. Assume that the line g is spanned by two points labeled 1 and 2 and that line h is spanned by two points labeled 3 and 4. Then we have g =1 2 and h =3 4. The quadratic form becomes p, 1 2 p, 3 4 =0. We may as well express this term as the product of two determinants [p, 1, 2][p, 3, 4] = 0. Each factor describes a linear condition on the point p. The product calculates the conjunction between the two expressions. Now, assume that we want to describe the set of conics that passes through for points 1,..., 4 in general position. Clearly, there are many conics that satisfy this condition. The corresponding system of linear equations consists of four equations in six variables. Hence the solution space will be twodimensional. One of these two degrees of freedom goes into the homogeneity of

3 10.1 Conic through five points Fig Bundles of conics though four points. Three degenerate special cases. the conic parameters. Therefore we have a bundle of geometric solutions with one degree of freedom. Figure 10.1 (left) illustrates such a bundle of conics. Among these conics there are three degenerate conics, each of them passing through a pair of lines spanned by the four points. In Figure 10.1 (right) these pairs of lines are marked by identical colors. They correspond to the following four quadratic forms: [p, 1, 2][p, 3, 4] = 0, [p, 1, 3][p, 2, 4] = 0, [p, 1, 4][p, 2, 3] = 0. A linear combination of two of these forms (say the last two) λ[p, 1, 3][p, 2, 4] + µ[p, 1, 4][p, 2, 3] = 0 generates again a quadratic form. The set of points p satisfying this equation forms again a conic. This conic passes through all four points 1,..., 4 since both summands vanish on these points. If λ and µ run trough all possible values we obtain all the conics in the bundle through the four points. Applying the technique of Plücker s µ (compare Section 6.3) we can adjust these values 1 4 q 3 2 Fig Constructing a conic through five points.

4 Conics and perspectivities such that the resulting conic passes through another given point q. For this we have to simply choose λ =[q, 1, 4][q, 2, 3]; µ = [q, 1, 3][q, 2, 4]. The resulting conic equation can be written as [q, 1, 4][q, 2, 3][p, 1, 3][p, 2, 4] [q, 1, 3][q, 2, 4][p, 1, 4][p, 2, 3] = 0. Observe that this equation is a multi-homogeneous bracket polynomial that is quadratic in each of the six involved points. Figure 10.2 illustrates the situation. We can also interpret it as a bracket condition encoding the (projectively invariant) property that six points 1,...,4, p, q are on a conic (compare Section 6.4 and Section 7.3). We will come back to this interpretation in the next Section. Before this we will give the procedure for calculating calculate the symmetric matrix for the conic through the five points 1, 2, 3, 4,q. We give it as a kind of simple computer program: 1: g 1 := 1 3; 2: g 2 := 2 4; 3: h 1 := 1 3; 4: h 2 := 2 4; 5: G := g 1 g T 2 ; 6: H := h 1 h T 2 ; 7: M := q T HqG q T GqH; 8: A := M + M T ; The matrix A assigned in the last line of the program contains the symmetrized matrix Conics and cross ratios Let us come back to the equation [q, 1, 4][q, 2, 3][p, 1, 3][p, 2, 4] [q, 1, 3][q, 2, 4][p, 1, 4][p, 2, 3] = 0, ( ) which characterizes whether six points are on a conic. First observe that this equation is highly symmetric. For each bracket in one term its complement (the bracket consisting of the other three letters) is in the other term. The symmetry becomes a bit more transparent if we rewrite the equation with new points labels:

5 10.2 Conics and cross ratios p q Fig Four points on a conic seen from other points of a conic. [A, B, C][A, Y, Z][X, B, Z][X, Y, C] [A, B, Z][A, Y, C][X, B, C][X, Y, Z] = 0. There is another important observation that we can make by rewriting equation ( ). We assume that the conic is non-degenerate and that none of the determinants vanishes. In this case we can rewrite ( ) to the form [q, 1, 4][q, 2, 3] [p, 1, 4][p, 2, 3] = [q, 1, 3][q, 2, 4] [p, 1, 3][p, 2, 4]. Both sides of the equation represent cross ratios. The left side is a cross ratio of the lines p1, p2, p3, p4 the right side of the equation is a cross ratio of the lines q1, q2, q3, q4. We abbreviate (1, 2; 3, 4) q := [q, 1, 4][q, 2, 3] [q, 1, 3][q, 2, 4]. This is the cross ratio of 1, 2, 3, 4 as seen from point q. Thus equation ( ) may be restated as (1, 2; 3, 4) q = (1, 2; 3, 4) p. Point p and point q see the points 1, 2, 3, 4 under the same cross ratio. The situation is shown in Figure 10.3 We summarize this in a theorem: Theorem Let 1, 2, 3, 4,pbe five points on a conic such that p is distinct from the other four points. Then the cross ratio (1, 2; 3, 4) p is independent of the special choice of p. We will later on see that this theorem is very closely related to the so called exterior angle theorem for circles which states that in a circle a fixed secant is seen from an arbitrary point on the circle under the same angle (modulo π).

6 Conics and perspectivities The last theorem enables us to speak of the cross ratio of four points on a fixed conic as long as no more than two of the points (1, 2, 3, 4) coincide and we can speak of a cross ratio at all. For this we simply chose an arbitrary point p that does not coincide with 1, 2, 3, 4 and take the cross ratio (1, 2; 3, 4) p. The theorem is useful under many aspects. In particular it is useful to parameterize classes of objects. We will investigate two of these applications. First assume that the points 1, 2, 3, 4 are fixed. The last theorem states that for a fixed conic C the value of (1, 2; 3, 4) p is invariant of the choice of p. Thus it can be considered as a characteristic number that singles out the specific conic C from all other conics through the four points. Thus we can take this number as a kind of coordinate for the conic within the one-dimensional bundle of conics through 1, 2, 3, 4. In fact if we do so the three special degenerate conics in this bundle (compare Figure 10.1) correspond to the values 0, 1 and. In the second application we fix the conic itself as well as the the position of the points 1, 2, 3. The point p may be an arbitrary point on the conic whose exact position is not relevant for the calculations as long as it dies not coincide with the other points. If point 4 takes all possible positions on the conic, then the value of (1, 2; 3, 4) p takes all possible values of R { }, since the line p, 4 takes all possible positions through p. Thus we can use the cross ratio (1, 2; 3, 4) p to characterize the position of 4 with respect to 1, 2, 3 on the conic. The three special values 0, 1 and are assumed when 4 is identical to 1, 3, 2, respectively. In this setup we may consider the conic itself as a model of the real projective line. The three points 1, 2, 3 above play the role of a projective basis on this line with respect to which we measure the cross ratio. In this model it is obvious that the topological structure of the real projective line is a circle Perspective generation of conics The considerations of the last section can be reversed in order to create conics by perspective bundles of lines. For this we consider the points p and q as centers of two bundles of lines that are projectively related to each other. Forming the intersections of corresponding lines from each bundle creates a locus of points that all have to lie on a single conic. To formalize this fact (in particular to deal with the special cases) we have to sharpen our notions on projective transformations slightly. Definition Let l 1 and l 2 be two distinct lines in L R and o P R not incident to l 1 or l 2. Furthermore let P l1 and P l2 be the sets of points on the two lines, respectively. The map τ: P l1 P l2 defined by τ(p) = meet(l 2, join(o, p)) is called a (point-)perspectivity. We furthermore use the term projective transformation from P l1 to P l2 in the following sense. We represent the points on P li ; i =1, 2 by a suitable linear combinations α i a i + β i b i. If τ: P l1 P l2 can be expressed as

7 10.3 Perspective generation of conics 169 Fig A point perspectivity and a line perspectivity. ( ) α1 τ( )= β 1 ( )( ) ( ) ab α1 α2 =, cd β 1 β 2 then we call τ a projective transformation. Theorem 5.1 established that harmonic maps are projective transformation. In Lemma 4.3 we proved that perspectivities are particular projective transformations. Dually we can also speak about perspectivities of bundles of lines. Definition Let p 1 and p 2 be two distinct points in P R and o L R not incident to p 1 or p 2. Furthermore let L p1 and L p2 be the sets of lines through the two points, respectively. The map τ: L p1 L p2 defined by τ(l) = join(p 2, meet(o, l)) is called a (line-)perspectivity. Figure 10.4 shows images for both types of perspectivities. We will also consider projective transformations τ : L p1 L p2 in the corresponding dual sense to point transformations. Again line-perspectivities are special projective transformations. Now, we will use Theorem 10.1 to prove the following fact. Theorem Let p and q be two distinct points in RP 2. Let L p and L q be the sets of all lines that pass through p and q, respectively. Let τ: L p L q be a projective transformation which is not a perspectivity. Then the points meet(l, τ(l)) are all points of a certain conic C. Proof. Let l 1,l 2,l 3,l 4 be four arbitrary lines from L p not through q. Consider the points a i = meet(l i,τ(l i )); i =1,..., 4. Since the two bundles of lines were related by a projective transformation the cross ratio (l 1,l 2 ; l 3,l 4 ) equals the cross ratio (τ(l 1 ),τ(l 2 ); τ(l 3 ),τ(l 4 )). This relation can be written as (a 1,a 2 ; a 3,a 4 ) p =(a 1,a 2 ; a 3,a 4 ) q. Hence the six points a 1,a 2,a 3,a 4, p, q lie on a conic. Since τ is not a perspectivity the points a 1,a 2,a 3 cannot be collinear. (Assume on the contrary that they lie on a line l. Then the image of an arbitrary fourth line l 4 must satisfy the relation (l 1,l 2 ; l 3,l 4 ) =

8 Conics and perspectivities Fig Generation of a conic by projective bundles. (τ(l 1 ),τ(l 2 ); τ(l 3 ),τ(l 4 )). Hence the intersections of l 4 with l and τ(l 4 ) with l must coincide. This means that τ is a perspectivity.) Thus the conic C uniquely defined by p, q, a 1,a 2,a 3 is non-degenerate. Since l 4 was chosen to be arbitrary all other intersections a = meet(l, τ(l)) must lie on C as well. Conversely, for any point a on C{p, q} there is a line l that joins p and a. The intersection of l and τ(l) must be on the conic. Thus this intersection must be point a. The last theorem gives us a nice procedure to explicitly generate a conic as a locus of points. The conic is determined by two points p and q and a projective transformation between the line bundles through these two points. For generation of the conic we take a free line from the bundle L p and let it sweep through the bundle. All intersections of l and τ(l) form the points of the conic. Dually if we have two lines l 1 and l 2 whose point sets are connected by a projective transformation τ we can consider a point p freely movable on l 1. The lines join(p, τ(p)) forms the set of tangents to a particular conic. Figure 10.5 shows two particularly simple (but still interesting) examples of this generation principle. On the right two bundles of lines are shown where the second one simply arises from shifting and rotating the first one (this a particularly simple projective transformation). The resulting generated conic, that comes from intersecting corresponding lines is a circle. This result could also be derived elementary by using the exterior angle theorem for circles. We will later on see that this theorem is highly related to our conic constructions. The second example shows two sets of equidistant points on two different lines (they are again related by a projective transformation). Joining corresponding lines yields the envelopes of a circle. One should compare these two pictures with Figure 10.4 in which pairs of objects were shown that were related by a perspectivity. This case is the degenerate limit case of the above construction. Remark The construction underlying Theorem 10.2 also demonstrates that the sets of points on a non-degenerate conic can be polynomially parameterized (in homogeneous coordinates). For this consider two points p, q

9 10.4 Transformations and conics 171 on the conic. And the two corresponding bundles of lines together with the corresponding projective transformation τ. We introduce a projective basis on each of the two bundles together with a suitable homogeneous coordinatisation (say we represent lines from the first bundle by λ p l 1 + µ p l 2 and points from the second bundle ( ) by ( λ q m )( 1 + µ ) q m 2.) The projective transformation ( ) τ λq ab λp ab can be written as = for a suitable matrix. Thus the µ q cd µ p cd points on the conic have homogeneous coordinates (tl 1 + (1 t)l 2 ) ((at + b(t 1))m 1 +(ct + d(t 1))m 2 ). Here t is a parameter that runs through all elements of R from to +. By this we get all points of the conic except for the one corresponding to t =. The above formula is simply a polynomial function. A similar statement is no longer true for curves of higher degree. In general they cannot be parameterized by rational or even polynomial functions Transformations and conics In this section we will deal with two types of transformations. Those who change the shape of a conic (there we will study how we can derive the equation of the transformed conic) and those that leave the conic invariant. For them we will have a look at the transformation group generated by these transformations. The first task is simple. What happens to a conic under a projective transformation τ: RP 2 RP 2. The transformation is best understood if we write the conic equation in matrix form p T Ap =0. Now assume that we apply a projective transformation τ that is expressible as multiplication by an invertible 3 3 matrix T. Thus a point p on the conic becomes transformed to a point Tp. Such a point should be in the transferred conic. This implies that the equation of the transformed conic is (T 1 p) T A(T 1 p). Thus we obtain the matrix of the transferred conic as T 1T AT 1. Analogously the equation of the dual conic l T Bl = 0 transfers to (T T l) T B(T T l) and the matrix of the transformed dual conic becomes T BT T.

10 Conics and perspectivities Let us turn to the more interesting task of studying all those projective transformations that leave a given fixed conic C invariant. Such a transformation must map points on C to points on C. We here discuss the non-degenerate case only and postpone the degenerate case to later chapters. The key to the classification of such transformations is Theorem 10.1 which allows us to identify the points on a conic with the points on a projective line and to associate a cross ratio to quadruples of such points. Our aim is to prove that a projective transformation τ: RP 2 RP 2 that leaves C invariant induces a projective transformations on C (considered as a projective line). For the following considerations we fix a non-degenearate conic C and identify it with the projective line. As indicated in Section 10.2 we will speak of the cross ratio (1, 2; 3, 4) C of four points on C which is (1, 2; 3, 4) p. for a suitably non-degenerate choice of p. Theorem Let τ: RP 2 RP 2 be a projective transformation that leaves C invariant. Then the restriction of τ to C is a projective transformation on C Proof. Let 0, 1 and three distinct points on C. The position of an arbitrary point x on C is uniquely determined by the value of the cross ratio (0, ; 1, x) C. Let τ: RP 2 RP 2 be a projective transformation that leaves C invariant. We will prove that the position of τ(x) is already defined by the positions of τ(0), τ(1) and τ( ) and that we have in particular (0, ; 1, x) C =(τ(0),τ( ); τ(1),τ(x)) C. For this let p on C be chosen such that p does not coincide with the points 0, 1,, or x. Then τ(p) will automatically not coincide with τ(0), τ(1), τ( ), or τ(x). Since τ is a projective transformation we have (0, ; 1, x) p =(τ(0),τ( ); τ(1),τ(x)) τ (p). The special choice of the position of p guarantees that the cross ratios are well defined. Now p as well as τ(p) are on C. The other four image points are also on C. Thus the above two cross ratios are the cross ratios are the cross ratios (0, ; 1, x) C on C and (τ(0),τ( ); τ(1),τ(x)) C on C. Thus these two cross ratios must be equal as claimed. This implies that the restriction of τ to C must be a projective transformation. The proof of the last theorem was algebraically simple but conceptually interesting. It relates a projective transformation on RP 2 that leaves C invariant to its action on C itself. With our concept of C representing the projective line we see that in this world τ induces nothing else but a 1-dimensional projective transformation. In the theorem it was crucial that the value of the cross ratio of four points seen from a fifth point p is independent of the choice of p. This allowed us to relate the image seen from p to the image seen form τ(p). We can also take the opposite define a projective transformation that leave C invariant by explicitly giving the images of four suitably chosen points on C.

11 10.4 Transformations and conics 173 d τ(d) τ(c) a c τ τ(b) τ(a) b Fig A transformation that leaves a conic invariant. Theorem Let a, b, c, d, and a, b, c, d be two quadruples of distinct points on a non-degenerate conic C such that (a, b; c, d) C =(a,b ; c,d ) C. Then there exists a unique projective transformation τ: RP 2 RP 2 with τ(a) =a, τ(b) =b, τ(c) =c, τ(d) =d which furthermore leaves C invariant. Proof. The transformation τ is uniquely determined by the pre-image points a, b, c, d and the image points a,b,c,d. Thus we only have to show that τ indeed leaves C invariant. Since a non-degenerate conic is uniquely determined by five points on it it suffices to prove that there exists one more point p on C whose image τ(p) is also on C. For this let p be an arbitrary point distinct from the points a, b, c, d. Thus we have (a,b ; c,d ) C =(a, b; c, d) C =(a, b; c, d) p = (τ(a),τ(b); τ(c),τ(d)) τ (p) =(a,b ; c,d ) τ (p). The third equation holds since τ is a projective transformation. The fact that (a,b ; c,d ) C =(a,b ; c,d ) τ (p) shows that τ(p) also must lie on the conic C. Figure 10.6 shows a circle before and after a projective transformation that leaves the circle invariant. The transformation τ: RP 2 RP 2 is determined by the image of the four red points. The position of the four image points cannot be chosen arbitrarily. They must have the same cross ratio with respect to the circle as the four pre-image points. In the situation shown in the picture the four points are in harmonic position with respect to the circle. The white points in the pre-image circle (left) map to the white points in the image circle (right). The lines and the central point indicate how the interior of the circle is distorted by τ. We can also make the relation of C to the projective line RP 1 more explicit and relate the points of C to the line bundle L p of lines through a point p C.

12 Conics and perspectivities Such a line bundle considered as a set of lines is by duality a representation of the projective line. We can explicitly relate every point on C to a line in L p : Each line is associated to its intersection with C different from p. There is one line in the bundle that has to be treated separately. The tangent through p is associated to p itself. (This reflects the limit situation when the point on C approaches p). We can express this relation by a bijective map φ p : C L p from C to the bundle of lines through p. Now the last theorem states that the projective transformation τ induces a projective transformation τ p : L p L p in this line bundle via: τ p (l) := φ p (τ(φ 1 p (l))). The reader is invited to convince himself that the limit case of the tangent through p fits seamlessly into this picture. Figure 10.7 illustrates the relation of the points on the conic to the line bundle. In addition to the line bundle the picture also shows an additional line l that is intersected with every line of the bundle. So the points on the conic are also in one-to-one correspondence to the points on l. The picture exemplifies also how this relation of points on the conic to points on the line is closely related to the classical stereographic projection, a relation that will become much more important later. It is kind of remarkable how important it is that the point p is really placed on the conic. If it were inside the conic we would get a two-to-one relation between points on the conics and lines in the bundle L p. It point p were outside the conic not all lines of the bundle would intersect the conic at all. An intersection of a line and a conic corresponds to solving a quadratic equation. The fact that we consider a bundle at a point on the conic implies that we already know one of the two solutions of this quadratic equation. Thus solving the quadratic equation in principle can be reduced to a linear problem by factoring out the already known solution. The linearity is the deeper reason why there is a one-to-one correspondence of C and the lines in the bundle. Let us close this section with a remark on the group structure of the set of those transformations that leave C invariant. Theorem 10.3 can be interpreted in the following way: The group of all projective transformations that leaves a non-degenerate conic invariant is isomorphic to the group of transformations of RP Hesses Übertragungsprinzip The last sections made it clear that we can identify a non-degenerate conic with a projective line. In this section we will go even one step further. We will demonstrate a way how one can interpret arbitrary lines and points of RP 2 by suitable objects of the projective line. This allows us to represent statements in the two-dimensional world of RP 2 by corresponding statements

13 10.5 Hesses Übertragungsprinzip 175 φ l p(c) φ p(d) φ p(h) φ p(e) φ p(f) φ p(g) φ p(h) φ p(i) φ p(j) d e f g h l c i b a p k j φ p(p) Fig Generation of a conic by projective bundles. of certain objects on the projective line. The idea of this translation goes back to an article of Otto Hesse from Hesse was mainly interested in questions of invariant theory and studied several ways to linearize objects if higher degree. In his works around 1866 he was interested in generalizing the concept of duality. Duality allows us to derive for every theorem of projective geometry a corresponding dual theorem just by applying a dictionary that translates point by line, line by point, intersection by meet, and so forth. In the same spirit Hesse formulated a principle that allowed it to derive a 1-dimensional theorem from any two-dimensional theorem of projective geometry. He coined his principle by the term Übertragungsprinzip. A reasonable translation of this term could be principle of transfer. His work had far reaching consequences. It was used by Klein in his famous Erlanger Programm to demonstrate the concept of equivalent geometries. It inspired further work and many interesting generalizations. Some of these generalizations had important impact on the classification of Lie algebras or even on quantum theory. Within the present book we will use the transfer principle for deriving elegant bracket expressions for geometric configurations involving conics and lines. In his original work Hesse related Points in RP 2 to solutions of onedimensional quadratic forms. We will take a slightly more visual approach that allows us to represent the solutions of the quadratic forms directly as intersections of a conic with a line. As before we consider a non-generate conic C as an image of a projective line. Now to a line l in RP 2 we associate its two points of intersection with C. A word of caution is necessary. First of all not all lines will have two intersections with C. This corresponds to the situation that Hesse studied solutions of arbitrary quadratic forms with real coefficients. There may be two real solutions, two complex solutions (which are conjugates) or one (double) real solution. The three cases correspond to the situations where the line intersects in two, in no or in one point, respectively. To state Hesse s ideas in full generality we have to also deal with the complex

14 Conics and perspectivities solutions. This will be our first careful investigation of complex situations in projective geometry. Thus to treat Hesses transfer principle properly we must talk about CP 1 instead of RP 1. However, the only objects we have to consider are pairs of points (p, q) which are either both real or complex conjugates (p = q) or coincide (p = q). For the following considerations one may either consider these complex elements (all algebraic considerations work straight forward) or one may assume (for convenience) that the conic is large enough such that all lines under consideration intersect it in at least one point. A line l that intersects the conic C in two (real or complex) points p 1, and p 2 is represented by the pair H C (l) := (p1,p2). If l is tangent to the conic at point p we represent it by the pair H C (l) := (p, p). In all our considerations related to Hesse s transfer principle the order of the points within such a pair will be irrelevant. Nevertheless it is important to speak of pairs rather than sets to cover also the situation of a double point (p, p). If lines are represented by pairs of points what is the corresponding representation of a point of RP 2? In Hesse s transfer principle points would be represented by projective transformations on the projective line that are furthermore involutions (i.e. τ 2 = id). Such a transformation is derived in the following way. For a point p not on C we take two arbitrary distinct lines l and m through p that intersect C and consider the pairs of points H C (l) = (a 1,a 2 ) and H C (m) = (b 1,b 2 ). These four points are distinct and since they lie on a nondegenerate conic no three of them are collinear. Thus there is a unique projective transformation τ: RP 2 RP 2 that simultaneously interchanges a 1 with a 2 and b 1 with b 2. In particular this transformation leaves l, m and p invariant. Furthermore we have (by Theorem 4.2) (a 1,a 2 ; b 1,b 2 ) C =(a 2,a 1 ; b 2,b 1 ) C. This in turn implies by Theorem 10.3 that τ leaves the conic C invariant. Such a projective transformation induces by Theorem 10.2 a corresponding transformation τ p on C considered as RP 1. This is the object to which p is translated. The crucial fact on the definition of τ p is that it only depends on the choice of p but it is independent of the particular choice of l and m. We will not do this here. The reason for this is that we want to bypass a certain technical problem related to expressing a point p by a projective transformation τ p. If the point p is on the conic C then the above construction does not lead to a proper projective transformation, since a 1 and b 1 (or b 2 ) become identical. Instead of introducing a concrete object that represents a point we will characterize concurrence of lines k, l, m directly by a relation of the corresponding point pairs H C (k), H C (l)and H C (m). This characterization also covers the degenerate cases in which the coincident point lies on C. Theorem Let C be a conic and let k, l, m be lines of RP 2. To exclude the complex case we assume that they intersect or touch the conic. If k, l, m are concurrent then (H C (k); H C (l); H C (m)) form a quadrilateral set. We will prove this theorem by restriction to a remarkable special case by a suitable projective transformation. This special case was communicated by

15 10.5 Hesses Übertragungsprinzip 177 l 3 H(l 1) = (q, q) l 1 H(l 1) = (p 1,p 2) p 2 l 2 p 1 H(l 1) = (p, p) p Fig Hesse s transfer principle for lines. Each line is associated to a pair of points. In case the line does not intersect the conic the points are complex and conjugates. Yuri Matiyasevich (private communication) who discovered this remarkable configuration as a high-school student. Matiyasevich s configuration is a kind of geometric gadget for performing multiplications. He used this gadget to give a geometric construction for the prime numbers. We formulate it in the real euclidean plane: Lemma Let x and y be two real numbers. The join of the points ( x, x 2 ) and (y, y 2 ) crosses the y-axis at the point (0,x y). Proof. We can proof this by direct calculation when we show that the three points are collinear. det x y 0 x 2 y 2 xy = x y 2 + y xy ( x) xy y x 2 = Figure 10.9 gives an impression of how the parabola-multiplication-device works. For our purposes we must also cover the degenerate case y = x. Then the join becomes a tangent and we obtain: Lemma Let x be a real number. Then the tangent at ( x, x 2 ) to the parabola y = x 2 crosses the y-axis at the point (0, x 2 ). Proof. Also this can easily checked by direct calculation. The tangent has slope 2x Hence the tangent has the equation f(t) =a 2x.t resolving for a gives x 2 = a 2x( x). Thus a must be x 2.

16 Conics and perspectivities Fig Multiplying by a parabola. Now, what has Matiyasevich s gadget to do with Hesse s transfer principle. The parabola plays the role of the conic. The points on the conic are vertically projected onto the x-axis. Thus the x-axis is the representation of RP 1 that is isomorphic to the points on the conic (the unique infinite point of the parabola corresponds to the point at infinity of the x-axis). The line shown in Figure 10.9 intersects the conic in two points (the green and the blue one). They are associated to their x-value by the projection. Thus the green and blue point on the x axis corresponds to the Hesse-pair that represents the line. Now we are ready to prove Theorem 10.5 (which is essentially Hesse s transfer principle) as a simple Application of Matiyasevich s construction. Proof of Theorem 10.5: Since three tangents of a conic C never intersect in one point at least one of the lines must meet the conic in two points. After a suitable projective transformation we may assume that the conic p is the parabola y = x 2 (in Euclidean coordinates) and that one of the lines (say k) is the y-axis. We identify the x-axis together with its point at infinity with the RP 1 associated to the conic. The corresponding mapping goes via vertical projection. Thus k is mapped to H C (k) = (0, ). Now assume that the other two lines l and m intersect the y-axis at the same point as required by the theorem. Let the corresponding point pairs on the x-axis be H C (l) = (l x,l y ) and H C (m) = (m x,m y ). Since the two lines in the theorem intersect the y-axis in the same point we can consider them as an two instances of Matiyasevich s construction and we get: ( l x ) (l y ) = ( m x ) (m y ). This expression can be used to prove the corresponding quadset relation. For this we introduce homogeneous coordinates ( ) ( ) ( ) ( ) ( ( lx ly mx my 0 1,,,,, ) 0)

17 10.5 Hesses Übertragungsprinzip 179 Fig Hesse s transfer principle as incidence theorem. and calculate the characteristic quadset equation of Section 8.2. For the six points (l x,l y ; m x,m y ;0, ) being a quadset we must show [l x, ][m x,l y ][0,m y ] = [l x,m y ][m x, ][0,l y ]. This expands to l x 1 10 m x l y m y 1 1 = l x m y 1 1 m x l y 11. Expanding the determinants yields: which reduces to ( 1)(m x l y )( m y ) = (l x m y )( 1)( ly) m x m y + l y m y = l x l y + m y ly. Subtracting l y m y on both sides leaves us exactly with the identity proved by Matiyasevich s equation. With Theorem 10.5 we reduced the essence of Hesse s transfer principle to an incidence theorem in the projective plane. Lines are represented by pairs of points. Three lines intersect if the corresponding three pairs of points form a quadrilateral set. Figure summarizes the essence of Hesse s transfer principle as an incidence theorem. The green lines are the three lines that intersect. The six points of intersection seen from one point on the boundary of the conic generate a line bundle that must form a quadrilateral set. The red part of the figure certifies the quadset relation by the construction given in Figure 8.2.

18 Conics and perspectivities X Y Fig Pascal s Theorem 10.6 Pascal s and Brianchon s Theorem No exposition on conics would be complete without a treatment of Pascal s Theorem. This theorem was discovered already in 1640 by the famous Blaise Pascal and can be considered as a generalization of Pappus s Theorem. Figure shows an instances of this theorem. Theorem If 1,..., 6 are six points on a conic then the three intersections of opposite sides the hexagon (1, 2, 3, 4, 5, 6) are collinear. Proof. We already presented proofs of this theorem in Chapter 1. However, this time we want to add another prove which is a simple application of Hesse s transfer principle. We may assume that the three intersection points are distinct, since otherwise they are trivially collinear. For the labeling refer to Figure In order to apply the transfer principle we will simply express the three inner intersections of Pascal s Theorem as quadrilateral set conditions. Since the six points 1,..., 6 all lie on the conic we can identify them (applying the transfer principle) with points in RP 1. We will also need two more points on the conic, namely the intersections X and Y with the central conclusion line. Also they are considered as points in RP 1. Now the fact that 12, 45, XY, meet in a point is equivalent to the condition that (1, 2; 4, 5; X, Y ) forms a quadrilateral set. This corresponds to the algebraic condition [1Y ][52][X4] = [14][5Y ][X2]. Similarly the fact that 34, 16, XY, meet in a point can be encoded by the equation: [3Y ][14][X6] = [36][1Y ][X4]. Multiplying both left and right sides and canceling brackets that appear on both (the distinctness of the intersection points implies that they are non-zero) sides leaves us with:

19 10.7 Harmonic points on a conic 181 Fig Brianchon s Theorem [52][3Y ][X6] = [5Y ][36][X2] which implies that 32, 56 and XY meet in a point and thus proves the theorem. For reasons of completeness we also mention the dual of Pascal s theorem. It is named after Charles Julien Brianchon and was discovered in 1804 (more than 150 years after Pascal s Theorem!). Theorem 10.7 (Brianchons Theorem). Let 1,..., 6 be six tangents to a conic (considered as the sides of a hexagon). Then the joins of opposite hexagon vertices meet in a point (see Figure 10.12). Pascal s Theorem also holds in limit cases in which one upto three consecutive points of the hexagon (1,..., 6) coincide. The join of two such consecutive points then becomes a tangent to the conic. We refer the reader to Section 1.4 for examples of such limit situations Harmonic points on a conic As a (for now) final application if Hesse s transfer principle we wan to show that it is extremely simple to construct a harmonic point on a non-degenerate conic. For this again we identify the conic C with the projective line. If three points a, b, c on C are given we want to construct a fourth point d such that (a, b; c, d) C = 1 holds. The construction us shown in Figure and just consists of two tangents at a and b and a join of their intersection to c. By Hesses transfer principle applied to this situation we get that (a, a; b, b; c, d) is a quadrilateral set (the tangents correspond to the double points(a, a) and (b, b)). This means that

20 Conics and perspectivities a c d b Fig Construction of a harmonic quadruple (a, b; c, d) = 1 [ab][bd][ca] = [ad][ba][cb]. Dividing one term by the other and canceling the bracket [a, b] gives: [bd][ca] [ad][cb] = 1. Which is after a slight reordering of the letters easily recognized as the condition for (a, b; c, d) being harmonic. It is an amazing fact that the construction of a harmonic point on a conic turns out to be even simpler than the corresponding task on a line. This reflects on the one hand the fundamental importance of conics and on the other hand the fact that conics are closely related to involutions and involutions are closely related to harmonic sets. In particular if we fix a and b and consider the construction of point d as a function τ: RP 1 RP 1 with τ(c) =d, then this map τ turns out to be a projective involution with fixpoints a and b.