A rigorous deductive approach of elementary euclidean geometry Jean-Pierre Demailly Université Joseph Fourier Grenoble I 0. Introduction The goal of this article is to explain a rigorous and still reasonably simple approach for teaching elementary Euclidean geometry at the secondary education levels. Euclidean geometry is a privileged area of mathematics, since it allows from an early stage to practice rigorous reasonings and to exercise vision and intuition. Our concern is that the successive reforms of curricula in the last 3 decades in France, and possibly in other western countries as well, have brought a worrying decline of geometry, along with a weakening of mathematical reasoning which geometry specifically contributed to in an essential way. We hope that these views will be of some interest to textbook authors and to teachers who have a possibility of not following too closely the prescriptions for weak contents, when they are unfortunately enforced by education authorities and curricula. The first sections should ideally also be mastered by primary school teachers, as it is always advisable to know more than what one has to teach at any given level~! 1. On axiomatic approaches of geometry As a formal discipline, geometry originates in Euclid's list of axioms, even though substantial geometric knowledge existed even before. An excerpt of Euclid's book The traditional teaching of geometry that took place in France during the period 1880-1970 was directly inspired by Euclid's axioms, stating first the basic properties of geometric objects and using the "triangle equality criteria" as the starting point of geometric reasoning. This approach had the advantage of being very effective and of quickly leading to rich contents. It also adequately reflected the intrinsic nature of geometric properties, without requiring extensive algebraic calculations. These choices echoed a mathematical tradition that was firmly rooted in the nineteenth century, aiming to develop "pure geometry", the highlight of which was the development of projective geometry by Poncelet. Euclid's axioms, however, were neither complete nor entirely satisfactory from a logical perspective, leading mathematicians as Hilbert and Pasch to develop the system of axioms now attributed to Hilbert, that was settled in his famous memoir Grundlagen der Geometrie in 1899. David Hilbert (1862-1943), in 1912 It should be observed, though, that the complexity of Hilbert's system of axioms makes it actually unpractical to teach geometry at an elementary level. The result, therefore, was than only a very partial axiomatic approach was taught, leading to a situation where a large number of properties that could have been proved formally had to be stated without proof, with the mere justification that they looked intutively true. This was not necessarily a major handicap, since pupils and their teachers may not even have noticed the logical gaps. However, such an approach, even though it was in some sense quite successful, meant that a substantial shift had to be accepted with more contemporary developments in mathematics, starting already with Descartes' introduction of analytic geometry. The drastic reforms implemented in France around 1970 (with the introduction of "modern mathematics") swept away all these concerns by implementing an entirely new paradigm~: according to Jean Dieudonné, one of the Bourbaki founders, geometry should be taught as a corollary of linear algebra, in a completely general and formal setting. The first step of the reform implemented this approach from "classe de seconde" (grade 10) on. A major problem, of course, is that the linear algebra viewpoint completely departs from the physical intuition of Euclidean space, where the group of invariance is the group of Euclidean motions and not the group of affine transformations. The reform could still be followed in a quite acceptable way for about one decade, as long as pupils had a solid background in elementary geometry from their earlier grades, but became more and more unpractical when primary school and junior high school curricula were themselves (quite unfortunately) downgraded. All mathematical contents of high school were then severely axed around 1985, resulting in curricula prescriptions that in fact did not allow any more the introduction of substantial deductive activity, at least in a systematic way. We believe however that it is necessary to introduce the basic language of mathematics, e.g. the basic concepts of sets, inclusion, intersection, etc, as soon as needed, most certainly already at the beginning of junior high school. Geometry is a very appropriate groundfield for using this language in a concrete way. 2. Geometry, numbers and arithmetic operations An important issue is the relation between geometry and numbers. Greek mathematicians already had the fundamental idea that ratios of lengths with a given unit length were in one to one correspondence with numbers~: in modern terms, there is a natural distance preserving bijection between points of a line and the set of real numbers. This viewpoint is of course not at all in contradiction with elementary education since measuring lengths in integer (and then decimal) values with a ruler is one of of the first important facts taught at primary school. However, at least in France, several reforms have put forward the extremely toxic idea that the emergence of electronic calculators would somehow free pupils from learning elementary arithmetic algorithms for addition, subtraction, multiplication and division, and that mastering magnitude orders and the "meaning" of arithmetic operations would be more than enough to understand society and even to pursue in science. The fact is that one cannot conceptually separate numbers from the operations that can be performed on them, and that mastering algorithms mentally and in written form is instrumental to realizing magnitude orders and the relation of numbers with physical quantities. The first contact that pupils will have with "elementary physics", again at primary school level, is probably through measuring lengths, areas, volumes, weights, densities, etc. Understanding the link with arithmetic operations is the basic knowledge that will be involved later to connect physics with mathematics. The idea of a real number as a possibly infinite decimal expansion then comes in a natural way when measuring a given physical quantity with greater and greater accuracy. Square roots are forced upon us by Pythagoras' theorem, and computing their numerical values is also a very good introduction to the concept of real number. I would certainly recommend to (re)introduce from the very start of junior high school (not later than grade 6 and 7), the observation that fractions of integers produce periodic decimal expansions, e.g. 1/7~=~0.142857142857..., while no visible period appears when computing the square root of 2. In order to understand this (and before any formal proof can be given, as they are conceptually harder to grasp), it is again useful to learn here the hand and paper algorithm for computing square roots, which is only slightly more involved than the division algorithm and makes it immediately clear that there is no reason the result has to be periodic – unfortunately, this algorithm is no longer taught in France since a long time. When all this work is correctly done, it becomes really possible to give a precise meaning to the concept of real number at junior high school – of course many more details have to be explained, such as the identification of proper and improper decimal expansions, e.g. 1~=~0.99999... , the natural order relation on such expansions, decimal approximations with at given accuracy, etc. In what follows, we propose an approach of geometry based on the assumption that pupils have a reasonable understanding of numbers, arithmetic operations and physical quantities from their primary school years – with consolidation about things such infinite decimal expansions and square roots in the first two years of junior high school~; this was certainly the situation that prevailed in France before 1970, but things have unfortunately changed for the worse since then. Our small experimental network of schools SLECC ("Savoir Lire Ecrire Compter Calculer"), which accepts random pupils and operates in random parts of the country, shows that such knowledge can still be reached today for a very large majority of pupils, provided appropriate curricula are enforced. For the others, our views will probably remain a bit utopistic, or will have to be delayed and postponed at a later stage. 3. First steps of the introduction of Euclidean geometry 3.1. Fondamental concepts The primitive concepts we are going to use freely are~: real numbers, with their properties already discussed above~; points and geometric objects as sets of points~: a point should be thought of as a geometric object with no extension, as can be represented with a sharp pencil~; a line or a curve are infinite sets of points (at this point, this is given only for intuition, but will not be needed formally)~; distances between points. Let us mention that the language of set theory has been for more than one century the universal language of mathematicians. Although excessive abstraction should be avoided at early stages, we feel that it is appropriate to introduce at the beginning of junior high school the useful concepts of sets, of inclusion, the notation x  A, operations on sets such as union, intersection and difference~; geometry and numbers already provide rich and concrete illustrations. A geometric figure is simply an ordered finite collection of points Aj and sets Sk (vertices, segments, circles, arcs, ...) Given two points A, B of the plane or of space, we denote by d(A, B) (or simply by AB) their distance, which is in general a positive number, equal to zero when the points A et B coincide~- concretely, this distance can be mesured with a ruler. A fundamental property of distances is~: 3.1.1. Triangular inequality. For any triple of points A,B,C, their mutual distances always satisfy the inequality AC  AB + BC, in other words the length of any side of a triangle is always at most equal to the sum of the lengths of the two other sides. Intuitive justification. Let us draw the height of the triangle joining vertex B to point H on the opposite side (AC). If H is located between A and C, we get AC = AH + HC~; on the other hand, if the triangle is not flat (i.e. if H  B), we have AH < AB et HC < BC (since the hypotenuse is longer than the right-angle sides in a right-angle triangle – this will be checked formally thanks to Pythagoras' theorem). If H is located outside of the segment [A, C], for instance beyond C, we already have AC < AH  AB, therefore AC < AB  AB + BC. This justification1 shows that the equality AC = AB + BC holds if and only if the points A, B, C are aligned with B located between A and C (in this case, we have H = B on the left part of the above figure). This leads to the following intrinsic definitions that rely on the concept of distance, and nothing more2. 3.1.2. Definitions (segments, lines, half-lines). (a) Given two points A, B in a plane or in space, the segment [A, B] of extremities A, B is the set of points M such that AM + MB = AB. (b) We say that three points A, B, C are aligned with B located between A and C if B  [A, C], and we say that they are aligned (without further specification) if one of the three points belongs to the segment determined by the two other points. (c) Given two distinct points A, B, the line (AB) is the set of points M that are aligned with A and B; the half-line [A, B) of origin A containing point B is the set of points M aligned with A and B such that either M is located between A and B, or B between A and M. Two half-lines with the same origin are said to be opposite if their union is a line. In the definition, part (a) admits the following physical interpretration~: a line segment can be realized by stretching a thin and light wire between two points A and B~: when the wire is stretched, the points M located between A and B cannot "deviate", otherwise the distance AB would be shorter than the length of the wire, and the latter could still be stretched further ... We next discuss the notion of an axis: this is a line D equipped with an origin O and a direction, which one can choose by specifying one of the two points located at unit instance from O, with the abscissas +1 and -1~; let us denote them respectively by I and V. A point M  [O, I) is represented by the real value xM = + OM and a point M on the opposite half-line [O, I ') = xM by the real value –~OM. The algebraic measure of a bipoint (A, B) of the axis is defined by AB = xB - xA, which is equal to + AB or –~AB according to whether the ordering of A, B corresponds to the orientation or to its opposite. For any three points A, B, C of D , we have the Chasles relation AB + BC = AC This relation can be derived from the equality (xB - xA) + (xC - xB) = (xC - xA) after a simplification of the algebraic expression. Building on the above concepts of distance, segments, lines and half-lines, we can now define rigorously what are planes, half-planes, circles, circle arcs, angles …3 3.1.3. Definitions. (a) Two lines D , D' are said to be concurrent if their intersection consists of exactly one point. (b) A plane P is a set of points that can be realized as the union of a family of lines (UV) such that U describes a line D and V a line D~', for some concurrent lines D and D~' in space. If A, B, C are 3 non aligned points, we denote by (ABC) the plane defined by the lines D = (AB) and D~'~= (AC) (say).4 (c) Two lines D et D' are said to be parallel if they coincide, or if they are both contained in a certain plane P and do not intersect. (d) A salient angle (or a salient angular sector) defined by two non opposite half-lines [A,~B), [A,~C) with the same origin is the set obtained as the union of the family of segments [U,~V] with U~~[A,B) and V ~~[A,C). (e) A reflex angle (or a reflex angular sector) is the complement of the corresponding salient angle in the plane (ABC), in which we agree to include the half-lines [A,~B) and [A,~C) in the boundary. (f) Given a line D and a point M outside D , the half-plane bounded by D~containing M is the union of the two angular sectors and obtaained by expressing D as the union of two opposite half-lines [A, B) and [A, C) ; this is the union of all segments [U,~V] such that U~~D and V~~[A,~M). The opposite half-plane is the one associated with the half-line [A,~M') opposite to [A,~M). In that situation, we also say that we have flat angles of vertex~A. (g) In a given plane P , a circle of center A and radius R > 0 is the set of points M in the plane P such that d(A, M) = AM = R. (h) A circular arc is the intersection of a circle with an angular sector, the vertex of which is the center of the circle. (i) The measure of an angle (in degrees) is proportional to the length of the circular arc that it intercepts on a circle whose center coincides with the vertex of the angle, in such a way that the full circle corresponds to 360°. A flat angle (cut by a half-plane bounded by a diameter of the circle) corresponds to an arc formed by a half-circle and has measure 180°. A right angle is one half of a flat angle, that is, an angle corresponding to the quarter of a circle, in other words, an angle of measure equal to 90°. (j) Two half-lines with the same origin are said to be perpendicular if they form a right angle5. The usual properties of parallel lines and of angles intercepted by such lines ("corresponding angles" vs "alternate angles") easily leads to establishing the value of the sum of angles in a triangle (and, from there, in a quadrilateral). Definition (i) requires of course a few comments. The first and most obvious comment is that one needs to define what is the length of a circular arc, or more generally of a curvilign arc~: this is the limit (or the upper bound) of the lengths of polygonal line inscribed in the curve, when the curve is divided into smaller and smaller portions (cf. 2.2)6. The second one is that the measure of an angle is independent of the radius R of the circle used to evaluate arc lengths; this follows from the fact that arc lengths are proportional to the radius~R, which itself follows from Thales' theorem (see below). Moreover, a proportionality argument yields the formula for the length of a circular arc located on a circle of radius~R : a full arc (360°) has length 2πR, hence the length of an arc of 1° is 360 times smaller, that is 2~π~R~/~360 = π~R~/~180, and an arc of measure a (in degrees) has length l = (π~R~/~180)  a = R ~a~~π / 180. 3.2. Construction with instruments and isometry criteria for triangles As soon as they are introduced, it is extremely important to illustrate geometric concepts with figures and construction activities with instruments. Basic constructions with ruler and compasses, such as midpoints, medians, bissectors, are of an elementary level and should be already taught at primary school. The step that follows immediately next consists of constructing perpendiculars and parallel lines passing through a given point. At the beginning of junior high school début du collège, it becomes possible to consider conceptually more advanced matters, e.g. The problem of constructing a triangle ABC with a given base BC and two other elements, for instance : (a) the lengths of sides AB and AC, (b) the measures of angles and , (c) the length of AB and the measure of angle . In the first case, the solution is obtained by constructing circles of centers B, C and radii equal to the given lengths AB and AC, in the second case a~protractor is used to draw two angular sectors with respective vertices B and C, in the third case one draws an angular sector of vertex B and a circle of center~B. In each case it can be seen that there are exactly two solutions, the second solution being obtained as a triangle A'BC that is symmetric of ABC with respect to the line~(BC) : One sees that the triangles ABC and A'BC have in each case sides with the same lengths. This leads to the important concept of isometric figures. 3.2.4. Definition. (a) One says that two triangles are isometric if the sides that are in correspondence have the same lengths, in such a way that if the first triangle has vertices A, B, C and the corresponding vertices of the second one are A', B', C', then A'B' = AB, B'C' = BC, C'A' = CA. (b) More generally, one says that two figures in a plane or in space are isometric, the first one being defined by points A1, A2, A3, A4 ... and the second one by corresponding points ,~... if all mutual distances coincide. The concept of isometric figures is related to the physical concept of solid body: a body is said tto be a solid if the mutual distances of its constituents (molecules, atoms) do not vary while the object is moved; after such a move, atoms which occupied certain positions Ai occupy new positions and we have . This leads to a rigorous definition of solid displacements, that have a meaning from the viewpoints of mathematics and physics as well. 3.2.5. Definition. Given a geometric figure (or a solid body in space) defined by characteristic points A1, A2, A3, A4 ..., a solid move is a continuous succession of positions Ai(t) of these points with respect to the time~t, in such a way that all distances Ai(t)Aj(t) are constant. If the points Ai were the initial positions and the points are the final positions, we say that the figure (~...) is obtained by a displacement of figure (A1 A2 A3 A4 ...).7 Beyond displacements, another way of producing isometric figures is to use a reflection (with respect to a line in a plane, or with respect to a plane in space, as obtained by taking the image of an object through reflection in a mirror)8. This fact is already observed with triangles, the use of transparent graph paper is then a good way of visualizing isometric triangles that cannot be superimposed by a displacement without "getting things out of the plane"~; in a similar way, it can be useful to construct elementary solid shapes (e.g. non regular tetrahedra) that cannot be superimposed by a solid move. 3.2.6. Exercise. In order to ensure that two quadrilaterals ABCD and A'B'C'D' are isometric, it is not sufficient to check that the four sides A'B' = AB, B'C = BC, CD' = CD, D'A' = DA possess equal lengths, one must also check that the two diagonals A'C = AC and B'D' = BD be equal~; equaling only one diagonal is not enough as shown by the following construction~: The construction problems considered above for triangles lead us to state the following fundamental isometry criteria. 3.2.7. Isometry criteria for triangles9. In order that two triangles be isometric, it is necessary and sufficient to check one of the following cases (a) that the three sides be respectively equal (this is just the definition), or (b) that they possess one angle with the same value and its adjacent sides equal, or (c) that they possess one side with the same length and its adjacent angles of equal values. One should observe that conditions (b) and (c) are not sufficient if the adjacency specification is omitted – and it would be good to introduce (or to let pupils perform) constructions demonstrating this fact. A use of isometry criteria in conjunction with properties of alternate or corresponding angles leads to the various usual characterizations of quadrilaterals – parallelograms, lozenges, rectangles, squares... 3.3. Pythagoras' theorem We first give the classical « Chinese » proof of Pythagoras' theorem, which is derived by a simple area argument based on moving four triangles (represented here in green, blue, yellow and light red). Its main advantage is to be visual and convincing10. The point is to compare, in the left hand and righ hand figures, the remaining grey area, which is the difference of the area of the square of side a + b with the area of the four rectangle triangles of sides a, b, c. The equality if the grey areas implies a² + b² = c² . Complement. Let (ABC) be a triangle and a, b, c the lengths of the sides that are opposite to vertices A,~B,~C. (i) If the angle is smaller than a right angle, we have c² < a² + b². (ii) If the angle is larger than a right angle, we have c² > a² + b². Proof. First consider the case where (ABC) is rectangle: we have c² = a² + b² and the angle is equal to 90°. We argue by either increasing or decreasing the angle ~: if angle is <~90°, we have c' < c~; if angle is >~90°, we have c" > c. By this reasoning, we conclude~: Converse of Pythagoras' theorem. With the above notation, if c² = a² + b², then must be a right angle, hence the given triangle is rectangle in C. 4. Cartesian coordinates in the plane The next fundamental stem of our approach is the introduction of cartesian coordinates and their use to give formal proofs of properties that had previously been taken for granted (or given with a partial justification only). This is done by working in orthonormal frames. 4.1. Expression of Euclidean distance Pythagoras' theorem shows that the hypotenuse MM' is given by the formula MM' ² = (x' – x)² + (y' – y)² , as the two sides of the right angle are x' – x et y' – y (up to sign). The distance from M to M' is given therefore by (It is of course advisable to first present the argument with simple numerical values). 4.2. Squares Let us consider the figure formed by points A(u; v), B (– v ; u), C (– u ; – v), D (v ; – u). Formula (4.1.1) yields AB² = BC² = CD² = DA² = (u + v)²+ (u – v) ² = 2(u² + v² ), hence the four sides have the same length, equal to . Similarly, we find OA = OB = OC = OD = , therefore the 4 isoceles triangles OAB, OBC, OCD and ODA are isometric, and as a consequence we have = 90° and the other angles are equal to 45°. Hence = 90°, and we have proved that our figure is a square. 4.3. « Horizontal and vertical » lines The set D of points M (x ; y) such that y = c (where c is a given numerical value) is a "horizontal" line. In fact, given any three points M, M', M" of abscissas x < x' < x" we have and therefore MM' + M'M" = MM". This implies by definition that our points M, M', M" are aligned. If we consider the line D1 given by the quation y = c1 with c1  c, this is another horizontal line, and we have clearly D~  D1 = , therefore our lines D~ and D1 are parallel. Similarly, the set D of points M (x ; y) such that x = c is a "vertical line" and the lines D~:~x~=~c, D1~:~~x =~c1 are parallel. 4.4. Line defined by an equation y = ax + b We start right away with the general case y = ax + b to avoid any repetitions, but with pupils it would be of course more appropriate to treat first the linear case y = ax. Consider three points M1 (x1~; y1), M2 (x2~; y2), M3 (x3~; y3) satisfying the relations y1~=~ax1~+~b, y2~=~ax2~+~b, y3~=~ax3~+~b, with x1~ < x2 < x3, say. As y2 – y1 = a( x2 – x1), we find and likewise - this shows that M1M2~+~M2M3 = M1M3, hence our points M1, M2, M3 are aligned. Moreover11, we see that for any point with , then this point is not aligned with M2 et M3, and similarly for such that . Consequence. The set D of points M (x ; y) such that y = ax + b is a line. The slope of line D is the ratio between the "height variation" and the "horizontal variation", that is, for two points M1 (x1~; y1), M2 (x2~; y2) of D the ratio A horizontal line is a line of slope a = 0. When the slope a becomes very large, the inclination of the line D becomess intuitively close to being vertical. We therefore agree that conviendra a vertical line has infinite slope. Such an infinite value will be denoted by the symbol  (without sign). Consider two distinct points M1 (x1~; y1), M2 (x2~; y2) – If x1~ x2, we see that there exists a unique line D : y = ax + b passing through M1 and M2 : its slope is given by and we infer b = y1~ — ax1~ = y2~ — ax2. If x1 = x2, the unique line D passing through M1 , M2 is the vertical line of equation x = x1. 4.5. Intersection of two lines defined by their equations Consider two lines D : y = ax + b and D' : y = a'x + b'. In order to find the intersection D~~D' we write y = ax + b = a'x + b', and get in this way (a' — a)x = – (b' – b). Therefore, if a~~a' there is a unique intersection point M (x ; y) such that The 'intersection of D with a vertical line D' : x = c is still unique, as we immediately find the solution x~=~c, y~=~ac+ b. From this discussion, we can conclude : Theorem. Two lines D and D' possessing distinct slopes a, a' have a unique intersection point: we say that they are concurrent lines. On the contrary, if a = a' and moreover b ~ b', there is no possible solution, hence D  D'~=~, our lines are distinct parallel lines. If a = a' and b = b', the lines D and D' are equal, and they are still considered as being parallel. Consequence 1. Two lines D and D' of slopes a, a' are parallel if and only if their slopes are equal (finite or infinite). Consequence 2. If D is parallel to D' and if D' is parallel to D", then D is parallel to D". Proof. In fact, if a = a' and a' = a", then a = a". We can finally prove « Euclid's parellel postulate » (in our approach, this is indeed a rather obvious theorem, and not a postulate !) Consequence 3. Given a line D and a point M0, there is a unique line D ' parallel to D that passes through M0 . Proof. In fact, if D has a slope a and if M0 (x0~; y0), we see that – for a = , the unique possible line is the line D ' of equation x = x0 , – for a  , the line D ' has an equation y = ax + b with b = y0 — ax0 , therefore D' is the line that is uniquely defined by the equation D' : y — y0 = a(x – x0). 4.6. Orthogonality condition for two lines Let us consider a line passing through the origin D : y = ax. Select a point M (u ; v) located on D, M  O, that is u  0. Then . We know that the point M' (u' ; v') = (—v ; u) is such that the lines D = (OM) and (OM') perpendicular, thanks to the construction of squares presented in section~4.2. Therefore, the slope of the line D' = (OM') perpendicular to D is given by if a  0. If a = 0, the line D coincides with the horizontal axis, its perpencular through O is the vertical axis of infinite slope. The formula is still true in that case if we agree that (let us repeat again that here  means an infinite non signed value). Consequence 1. Two lines D and D' of slopes a, a' are perpendicular if and only if their slopes satisfy the condition (agreeing that and ). Consequence 2. If D  D' and D'  D" then D and D" are parallel. Proof. In fact, the slopes satisfy hence , and so a" = a. □ 4.7. Thales' theorem We start by stating a « Euclidean version" of the theorem, involving ratios of distances rather than ratios of algebraic measures. Thales' theorem. Consider two concurrent lines D, D' intersecting in a point O, and two prallel lines Δ1, Δ2 that intersect D in points A, B, and D' in points A', B'~; we assume that A,B, A', B' are different from O. Then the length ratios satisfy Proof. We argue by means of a coordinate calculation, in an orthonormal frame Oxy such that Ox is perpendicular to lines Δ1, Δ2, and Oy is parallel to lines Δ1, Δ2. In these coordinates, lines Δ1, Δ2 are "vertical" lines of respective equations Δ1 : x = c1, Δ2:~x~=~c2 with c1,~c2  0, and our lines D, D' admit respective equations D : y = ax, D'~:y~=~a'x. Therefore A (c1~; ac1), B (c2~; ac2), A' (c1~; a'c1), B' (c2 ~; a' c2). By Pythagoras' theorem we infer (after taking absolute values) : We have a'  a since D and D' are concurrent by our assumption, hence a' – a  0, and we then conclude easily that In a more precise manner, if we choose orientations on D, D' so as to turn them into axes, and also an orientation on Δ1 and Δ2 , we see that in fact we have an equality of algebraic measures Converse of Thales' theorem. Let D, D' be concurrent lines intersecting in~O. If Δ1 intersects D, D' in distinct points A, A'~, and Δ2 intersects D, D' in distinct points B, B' and if then Δ1 et Δ2 are parallel. Proof. It is easily obtained by considering the line δ2 parallel to Δ1 that passes through B, and its intersection point β' with D'. We then see that , hence β'~=~B' and δ2~=~Δ2, and as a consequence Δ2 ~=~δ2~// Δ1. □ 4.8. Consequences of Thales and Pythagoras theorems The conjunction of isometry criteria for triangles and Thales and Pythagoras theorems already allows (in a very classical way~!) to establish many basic theorems of elementary geometry. An important concept in this respect is the concept of similitude. Definition. Two figures (A1 A2 A3 A4 ...) and (~…) are said to be similar in the ratio k (k > 0) if we have for all segments and that are in correspondence. An important case where similar figures are obtained is by applying a homothety with a given center, say point O : if O is chosen as the origin of coordinates and if to each point M (x~;~y) we associate the point M' (x'~;~y') such that x' = kx, y'~=~ky, then formula (4.1.1) shows that we indeed have A'B' = |k| AB, hence by assigning to each point Ai the corresponding point we obtain similar figures in the ratio |k| ; this situation is described by saying that we have homothetic figures in the ratio~k ; this ratio can be positive or negative (for instance, if k = –~1, this is a central symmetry with respect to~O). The isometry criteria for triangles immediately extend into criteria for similarity. Similarity criteria for triangles. In order to conclude that two triangles are similar, ii is necessary and sufficient that one of the following conditions is met~: (a) the corresponding three sides are proportional in a certain ratio k > 0 (this is the definition); (b) the triangles have a corresponding equal angle and the adjacent sides are proportional~; (c) the triangles have two equal angles in correspondence. An interesting application of the similarity criteria consists in stating and proving the basic metric relations in rectangle triangles~: if the triangle ABC is rectangle in A and if H is the foot of the altitude drawn from vertex A, we have the basic relations AB² = BH  BC, AC² = CH  CB, AH² = BH  CH, AB  AC = AH  BC. In fact (for example) the similarity of rectangle triangles ABH and ABC leads to the equality of ratios One is also led in a natural way to the definition of sine, cosine and tangent of an acute angle in a rectangle triangle. Definition. Consider a triangle ABC that is rectangle in A. On defines In fact, the ratios only depend on the angle (which also determines uniquely the complementary angle = 90° – ), since rectangle triangles that share a common angle else than their right angle are always similar by criterion (c). Pythagoras' theorem then quickly leads to computing the values of cos, sin, tan for angles with "remarkable~values" 0°, 30°, 45°, 60°, 90°. 4.9. Computing areas and volumes It is possible – and therefore probably desirable - to justify many basic formulas concerning areas and volumes of usual shapes and solid bodies (cylinders, pyramids, cones, spheres), just by using Thales and Pythagoras theorems, combined with elementary geometric arguments12. We give here some indication on such techniques, in the case of cones and spheres. The arguments are close to those developped by Archimedes more than two centuries BC (except that we take here the liberty of reformulating them in modern algebraic notations). The volume of a cone with an arbitrary plane base of area A and height h is given by (4.9.1) One can indeed argue by a dilation argument that the volume V is proportional to h, and one also shows that is is proportional to A by approximating the base with a union of small squares. The proof is then reduced to the case of an oblique pyramid (i.e. to the case when the base is a rectangle). The coefficient is justified by observing that a cube can be divided in three identical oblique pyramids, whose summit is one of the vertices of the cube and the bases are the 3 adjacent opposite faces. The altitude of these pyramids is equal to the side of the cube, and their volume is thus of the volume of the cube. Archimedes formula for the area of a sphere. Since any two spheres of the same radius are isometric, their area depends only on the radius~R. Let us take the center O of the sphere as the origin, and consider the "vertical" cylinder of radius R tangent to the sphere along the equateur, and more precisely, the portion of cylinder located between the "horizontal" planes z = – R and z~=~R. We use a "projection" of the sphere to the cylinder : for each point M of the sphere, we consider the point M' on the cylinder which is the intersection of the cylinder with the horizontal line Dm passing by M and intersection the Oz axis. This projection is actually one of the simplest possible cartographic representations of the Earth. After cutting the cylinder along a meridian (say the meridian of longitude 180°), and unrolling the cylinder into a rectangle, we obtain the following cartographic map. We are going to check that the cylindrical projection preserves areas, hence that the area of the sphere is equal to that of the corresponding rectangular map of sides 2R et 2 π R : 4.9.2) A = 2R  2πR =~ 4~π R². In order to check that the areas are equal, we consider a « rectangular field" delimited by parellel and meridian lines, of very small size with respect to the sphere, in such a way that it can be seen as a planar surface, i.e. to a rectangle (for instance, on Earth, one certainly does not realize the rotundity of the globe when the size of the field does not exceed a few hundred meters). Let a, b be the side lengths of our "rectangular field", respectively along parallel lines direction and meridian lines direction, and a', b' the side lengths of the corresponding rectangle projected on the tangent cylinder. In the view from above, Thales' theoreme immediately implies In the lateral view, the two triangles represented in green are homothetic (they share a common angle, as the adjacent sides are perpendicular to each other). If we apply again Thales' theorem to the tangent triangle and more specifically to the sides adjacent to the common angle, we get The product of these equalities yields We conclude from there that the rectangle areas a  b and a'  b' are equal. This implies that the cylindrical projection preserves areas, and formula (4.9.2) follows. 5. An axiomatic approach of Euclidian geometry Although we have been able to follow a deductive presentation when it is compared to some of the more traditional approaches - almost all of the statements were "proven" from the definitions - it should nevertheless be observed that some proofs relied merely on intuitive facts – this was for instance the case of the "proof" of Pythagoras' Theorem. The only way to break the vicious circle is to take some of the facts that we feel necessary to use as "axioms", that is to say, to consider them as assumptions from which we first deduct all other properties by logical deduction~; a choice of other assumptions as our initial premises leads to non-Euclidean geometries (see section~10). As we shall see, the notion of a Euclidean plane can be defined using a single axiom, essentially equivalent to the conjunction of Pythagoras' Theorem - which was only partially justified - and the existence of Cartesian coordinates – which we had not discussed either. In case the idea of ​​using an axiomatic approach would look frightening, we want to stress that this section may be omitted altogether - provided pupils are somehow led to understand that the coordinate systems can be changed (translated, rotated, etc.) as needed. 5.1. The "Pythagoras / Descartes" model In our vision, plane Euclidian geometry is based on the following "axiomatic definition ». Definition. What we will call a Euclidian plane is a set of points denoted , for which mutual distances of points are supposed to be known, i.e. there is a predefined function and we assume that there exist "orthonormal coordinate systems" : to each point one can assign a pair of coordinates , by means of a one-to-one correspon­dence satisfying the axiom13 (Pythagoras / Descartes) d(M, M') = for all points M (x ; y) and M' (x' ; y'). It is certainly a good practice to represent the choice of an orthonomal coordinate system by using a transparent sheet of graph paper and placing it over the paper sheet that contains the working area of the Euclidean plane (here that area contains two triangles depicted in blue, above which the transparent sheet of graph paper has been placed). This already shows (at an intuitive level only at this point) that there is an infinite number of possible choices for the coordinate systems. We now investigate this in more detail. 5.1.1. Rotating the sheet of graph paper around O by 180° A rotation of 180° of the graph paper around O has the effect of just changing the orientation of axes. The new coordinates (X ; Y) are given with respect to the old ones by X = – x, Y = – y . Since (– u)² = u² for every real number u, we see that the formula (*) d(M, M') = is still valid in the new coordinates, assuming it was valid in the original coordinates (x ; y). 5.1.2. Reversing the sheet of graph paper along one axis If we reverse along Ox, we get X = – x, Y = y and formula (*) is still true. The argument is similar when reversing the sheet along Oy. 5.1.3. Change of origin Here we replace the origin O by an arbitrary point M0 (x0~; y0) The new coordinates of point M (x ; y) are given by X = x–x0, Y = y – y0. For any two points M, M', we get in this situation X' - X = (x' - x0) - (x - x0) = x' – x, Y' – Y = (y' - y0) - (y - y0) = y' – y, and we see that formula (*) is still unchanged. 5.1.4. Rotation of axes We will show that when the origin O is chosen, one can get the half-line Ox to pass through an aribrary point M1 (x1 ; y1) distinct from O. This is intuitively obvious by "rotating" the sheet of graph paper around point O, but requires a formal proof relying on our "Pythagoras / Descartes" axiom. This proof is substantially more involved than what we have done yet, and can probably be jumped over at first - we give it here to show that there is no logical flaw in our approach. We start from the algebraic equality called Lagrange's identity (au + bv)² + (– bu + av)² = a² u² + b² v² + b² u² + a² v² = (a² + b²)(u² + v²), which is valid for all real numbers a, b, u, v. It can be obtained by developping the squares on the left and observing that the double products annihilate. As a consequence, if a and b satisfy a²~+~b²~=~1 (such an example is a = 3/5, b = 4/5) and if we perform the change of coordinates X = ax + by~, Y = – bx + ay we get, for any two points M, M' in the plane X' - X = a(x' – x) + b(y' – y), Y' -Y = – b(x' – x) + a(y' – y), (X' – X )² + (Y' - Y)² = (x' – x )² + (y' - y)² by Lagrange's identity with u = x' – x, v = y' – y. On the other hand, it is easy to check that aX - bY = x, bX + aY = y, hence the assignment is one-to-one. We infer from there that in the sense of our definition, (X ; Y) is indeed an orthonormal coordinate system. If we now choose a~=~kx1, b = ky1, the coordinates of point M1 (x1; y1) are transformed into X1 = ax1 + by1 = k (x1 + y1), Y1 = – bx1 + ay1 = k ( – y1x1 + x1y1) = 0, and the condition is satisfied by taking . Since and Y1 = 0, the point M1 is actually located on the half-line OX in the new coordinate system. 5.2. Revisiting the triangular inequality The proof given in 3.1.1, which relied on facts that were not entirely settled, can now be made completely rigorous. Given three distinct points A, B, C distincts, we select O = A as the origin and the half line [A,~C) as the Ox axis. Our three points then have coordinates A(0;0), B(u;v), C(c;0), c > 0, and the foot H of the altitude staring at B is H (u; 0). We find AC = c and Therefore AC = c = u + (c – u)  AB + BC in all cases. The equality only holds when we have at the same time v = 0, u~~0 and c~–~u~~0, i.e. u~~[0, c] and v = 0, in other words when B is located on the segment [A,~C] of the Ox axis. 5.3. Axioms of higher dimensional affine spaces The approach that we have described is also appropriate for the introduction of Euclidean geometry in any dimension, especially in dimension 3. The starting point is the calculation of the diagonal δ of a rectangular parallelepiped with sides a, b, c~: As the triangles ACD et ABC are ectangle in C and B respectively, we have AD² = AC² + CD² et AC² = AB² + BC² hence the « great diagonal » of our rectangle parallelepiped rectangle is given by = AD² = AB² + BC² + CD² =a² + b² + c² . This leads to the expression of the distance function in dimension 3 d(M, M') = and we can just adopt the latter in the 3-dimensional Pythagoras / Descartes axiom. 6. Foundations of vector calculus We will work here in the plane to simplify the exposition, but the only change in higher dimension would be the appearance of additional coordinates. 6.1. Median formula Consider points A, B with coordinates (xA~; yA), (xB~; yB) in an orthonormal frame Oxy.The point I of coordinates satisfies IA = IB = AB : this is the midpoint of segment [A, B]. Median formula. For every point M (x ; y), one has MA² + MB² = 2 MI² + AB² = 2 MI² + 2 IA². Proof. In fact, by expanding the squares, we get , while The median formula is obtained by adding the analogous equality for coordinates y and applying Pytahgoras' theorem. It follows from the median formula that there is a unique point M such that MA = MB = AB, in fact we then find MI² = 0, hence M = I. The coordinate formulas that we initially gave to define midpoints are therefore independent of the choice of coordinates. 6.2. Parallelograms A quadrilateral ABCD is a parallelogram if and only if its diagonals [A,~C] and [B,~D] intersect at their midpoint~: In this way, we find the necessary and sufficient condition which is equivalent to xB + xD = xA + xC, yB + yD = yA + yC, or, alternatively, to xB – xA = xC – xD, yB – yA = yC – yD, in other words, the variation of coordinates involved in getting from A to B is the same as the one involved in getting from D to C. 6.3. Vectors A bipoint is an ordered pair (A, B) of points; we say that A is the origin and that B is the extremity of the bipoint. The bipoints (A, B) et (A', B') are said to be equipollent if the quadrilateral ABB'A' is a parallelogram (which can possibly be a "flat" parallelogram in case the four points are aligned). Definition. Given two points A, B, the vector is the "variation of position" needed for getting from A to B. Given a coordinate frame Oxy, this "variation of position" is expressed along the Ox axis by xB – xA and along the Oy axis by yB – yA. If the bipoints (A~, B) and (A'~, B') are equipollent, the vectors and are equal since the variations and are the same (this is true in any coordinate system). The "components" of vector in the coordinate system Oxy are the numbers denoted in the form of an ordered pair (xB – xA~;~yB – yA). The components (s~;~t) of a vector depend of course on the choice of the coordinate frame Oxy~: to a given vector one assigns different components (s~;~t), (s'~;~t') in different coordinate frames Oxy, Ox'y'. 6.4. Addition of vectors The addition of vectors is defined by means of Chasles' relation (6.4.1) for any three points A, B, C~: when one takes the sum of the variation of position required to get from A to B, and then from B to C, one finds the variation of position to get from A to C~; actually, we have for instance (xB – xC) + (xB – xA)~=~xC – xC . Equivalently, if ABCD is a parallelogram, one can also put (6.4.2) That (6.4.1) et (6.4.2) are equivalent follows from the fact that in parallelogram ABCD. For any choice of coordintae frame Oxy, the sum of vectors of components (s~;~t), (s'~;~t') has components (s~+~s'~;~t~+~t'). For every point A, the vector has zero componaents~: it will be denoted simply . Obviously, we have for every vector . On the other hand, Chasles' relation yields for any two points A, B. Therefore we define in other words, the opposite of a vector is obtained by exchanging the origin and extremity of any corresponding bipoint. 6.5. Multiplication of a vector by a real number Given a vector of components (s~;~t) in a coordintae frame Oxy and an arbitrary real number , we define as the vector of components (s~;~t). This definition is actually independent of the coordinate frame Oxy. In fact if and ~~0, we have where C is the unique point located on the half-line [A, B) tsuch that AC = ~AB. On the other hand, if ~~0, we have –~~~0 and Finally, it is clear that . Multiplication of vectors by a number is distributive with respect to the addition of vectors (this is a consequence of the distributivity of multiplication with respect to addition in the set of real numbers). 7. Cartesian equation of circles and trigonometric functions By Pythagoras' thorem, the circle of center A (a, b) and radius Rin the plane is the set of points M satisfying the equation which can also be put in the form x²~+ y²~ –~ 2ax – 2by + c = 0 with c = a²~+ b²~– R². Conversely, the set of solutions of such an equation defines a circle of center A (a, b) and of radius if c < a²~+ b², reduced to point A if c = a²~+ b², and empty if c > a²~+ b² . The trigonometric circle C is defined to be the unit circle centered at the origin in an orthormal coordinate system Oxy, that is, the of points M~(x~; y) such that x²~+ y²~=~1. Let U be the point of coordinates (1~; 0) and V the point of coordinates (0~; 1). The usual trigonometric functions cos, sin and tan an then defined for arbitrary angle arguments as shown on the above figure14. The equation of the circle implies the relation (cos θ)² + (sin θ)² = 1 for every~θ. 8. Intersection of lines and circles Let us begin by intersecting a circle C of center A and radius R with an arbitrary line D. In order to simplify the calculation, we take A = O as the origin and we take to axis Ox to be perpendicular to the line D. The line D is then «~vertical~» in the oordinate frame Oxy. (We start here right away with the most general case, but, once again, it would be desirable to approach the question by treating first simple numerical examples...) This leads to equations C : x² + y² = R², D : x = x0, hence As a consequence, if |x0| < R, we have and there are two solutions and , corresponding to two intersection points and that are symmetric with respect to the Ox axis. If |x0|~=~R, we find a single solution y = 0~: the line D : x = x0 is tangent to circle C at point (x0~;~0). If |x0| > R, the equation y²~=~ R²~ – x²~ <~0 has no solution ; the line D does not intersect the circle. Consider now the intersection of a circle C of center A and radius R with a circle C ' of center A' and radius R'. Let d = AA' be the distance between centers. If d = 0 the circles are concentric and the discussion is easy (the circles coincid if R = R', and are disjoint if R~~R'). We will therefore assume that A~~A', i.e. d > 0. By selecting O = A as the origin and Ox = [A, A') as the positive x axis, we are reduces to the case where A (0 ; 0) and A' (d~; 0). We then get equations For any point M in the intersection C~~C ' , we thus get , hence This shows that the intersection C~~C ' is contained in the intersection C~~D of C~ with the line D : x = x0. Conversely, one sees that if x²~+ y² = R² and x = x0, then (x ; y) also satisfies the equation which is the equation of C ' , hence C~~D ~~C~~C ' and finally C~~D ~=~C~~C ' . The intersection points are thus given by . As a consequene, we have exactly two solutions that are symmetric with respect to the line (AA') as soon as –~R < x0 < R, or equivalently i.e. |R~–~R'| < d < R + R'. If one of the inequalities is an equality, we get x0~=~±R and we thus find a single solution y = 0. The circles are tangent internally if d~=~|R~–~R'| and tangent externally if d~=~R~+~R'. Note that these results lead to a complete and rigorous proof of the isometry criteria for triangles~: up to an orthonormal change of coordinates, each of the three cases entirely determines the coordinates of the triangles modulo a reflection with respect to Ox (in this argument, the origin O is chosen as one of the vertices and the axis Ox is taken to be the direction of a side of known length). The triangles specified in that way are thus isometric. 9. Scalar product The norm |||| of a vector is the length AB = d(A, B) of an arbitrary bipoint that defines . From there, we put (9.1) in particular . The real number is called the scalar product of and , and is also defined to be the scalar square of , denoted . Consequently we obtain By definition (9.1), we have (9.2) and this formula can also be rewritten (9.2') This was the main motivation of the definition~: that the usual identity for the square of a sum be valid for scalar products. In dimension 2 et in an orthonormal frame Oxy, we find ~; if has components (x' ; y'), definition (9.1) implies (9.3) In dimension n, we would find similarly From there, we derive that the scalar product is "bilinear", namely that If , are two vectors, we can pick a point A and write , then , so that . The triangle ABC is rectangle if and only if we have Pythagoras' relation AC²~=~AB²~+~BC², i.e. in other words, by (9.2), if and only if . Consequence. Two vectors and are perpendicular if and only if . More generall, if we fix an origin O and a point A such that , one can also pick a coordinate system such that A belongs to the Ox axis, that is, A = (u~;~0). For every vector in Oxy, we then get whereas As the half-line [O, B) intersects the trigonometric circle at point (kv ; kw) with we get by definition This leads to the very useful formulas 10. More advanced material At this point, we have all the necessary foundations, and the succession of concepts to be introduced becomes much more flexible - much of what we discuss below only concerns high school level and beyond. One can for example study further properties of triangles and circles, and gradually introduce the main geometric transformations (in the plane to start with)~: translations, homotheties, affinities, axial symmetries, projections, rotations with respect to a point~; and in space, symmetries with respect to a point, a line or a plane, orthogonal projections on a plane or on a line, rotation around an axis. Available tools allow making either intrinsic geometric reasonings (with angles, distances, similarity ratios, ...), or calculations in Cartesian coordinates. It is actually desirable that these techniques remain intimately connected, as this is common practice in contemporary mathematics (the period that we describe as "contemporary" actually going back to several centuries for mathematicians, engineers, physicists. ..) It is then time to investigate the phenomenon of linearity, independently of any distance consideration. This leads to the concepts of linear combinations of vectors, linear dependence and independence, non orthonormal frames, etc,~in relation with the resolution of systems of linear equations. On is quickly led to determinants 2~~2, 3~~3, to equations of lines, planes, etc. The general concept of vector space provides an intrinsic vision of linear algebra, and one can introduce general affine spaces, bilinear symmetric forms, Euclidean and Hermitian geometry in arbitrary dimension. What we have done before can be deepened in various ways, especially by studying the general concept of isometry. Définition. Let and be two Euclidean spaces and let ~be an arbitrary map between these. We say that s is an isometry from to if for every pair of points (M, N) of ~, we have d(s(M), s(N)) = d(M, N). Isometries are closely tied to scalar product via the following fundamental theorem. Theorem. If is an isometry, then s is an affine transform, and its associated linear map is an orthogonal transform of Euclidean vector spaces, namely a linear map preserving orthogonality and scalar products~: for all vectors . In the same vein, one can prove the following result, which provides a rigorous mathematical justification to all definitions and physical considerations appeared in section 3.2. Theorem. Let and be two isometric figures formed by points of a Euclidean space~. Then there exists an isometry s of the entire Euclidien space such that . Non Euclidean geometries. Bernhard Riemann (1826-1866) In contemporary mathematics, non Euclidean geometries are best seen as a special instance of Riemannian geometry, so called in reference to Bernhard Riemann, one of the founders of modern complex analysis and differential geometry. A Riemannian manifold is by definition a differential manifold M, namely a topological space that admits local differentiable systems of coordinates x~=~(x1~;~x2~;~...~; xn), equipped with an infinitesimal metric of the form By integration the infinitesimal metric along paths, one obtains the geodesic distance which is used as a substitute of the Euclidean distance (in physics, general relativity also arises in a similar way by considering Lorentz-like metrics of the form ). On the unit disk in the complex plane, denoting z = x + iy, one considers the so-called Poincaré metric (named after Henri Poincaré, 1854–1912) The associated geodesic distance can be computed to be When substituting this distance to the Pythagoras / Descartes axiom, one actually obtains a non Euclidean geometry, which is a model of the hyperbolic geometry discovered by Nikolai Lobachevski (1793-1856). In this geometry, there are actually infinitely many parallel lines to a given line through a given point p exterior to~, so that Euclid's fifth postulate fails~! Lobachevski's hyperbolic geometry and the failure of Euclid's fifth postulate On some ideas of Felix Hausdorff and Mikhail Gromov. We first describe a few important ideas due to Felix Hausdorff (1868~–~1942), one of the founders of modern topology. The first one is that the Lebesgue measure of can be generalized without any reference to the vector space structure, but just by using the metric. If (, d) is an arbitrary metric space, one defines the p‑dimensional Hausdorff measure of a subset A of as where est least upper bound of sums running over all countable partitions d with . For p~=~1 (resp. p~=~2, p~=~3) one recovers the usual concepts of length, area, volume, and the definition even works when p is not an integer (it is then extremely useful to define the dimension of fractal sets). Another important idea of Hausdorff is the existence of a natural metric structure on the set of compact subsets of a given metric space (, d). If K, L are two compact subsets of , the Hausdorff distance of K and L is defined to be The study of metric structures has become today one of the most active domains in mathematics. We should mention here the work of Mikhail Gromov (Abel prize 2009) on length spaces and "moduli spaces" of Riemannian manifolds. If X and Y are two compact metric spaces, one defines their Gromov-Hausdorff distance to be the infimum of all Hausdorff distances for all possible isometric embeddings , of X and Y in another compact metric space. This provides a crucial tool to study deformations and degenerations of Riemannian manifolds.