VOL. 76, NO.

1, FEBRUARY 2003

15

A Tale of Three Circles

C H A R L E S I. D E L M A N GREGORY GALPERIN
Eastern Illinois University Charleston, IL 61920 [email protected], [email protected]

Everyone knows that the sum of the angles of a triangle formed by three lines in the plane is 180 , but is this still true for curvilinear triangles formed by the arcs of three circles in the plane? We invite the reader to experiment enough to see that the angle sum indeed depends on the triangle, and that no general pattern is obvious. We give a complete analysis of the situation, showing along the way, we hope, what insights can be gained by approaching the problem from several points of view and at several levels of abstraction. We begin with an elementary solution using only the most basic concepts of Euclidean geometry. While it is direct and very short, this solution is not complete, since it works only in a special case. The key to another special case turns out to be a model of hyperbolic geometry, leading us to suspect that the various manifestations of the problem lie on a continuum of models of geometries with varying curvature. This larger geometric framework reveals many beautiful unifying themes and provides a single method of proof that completely solves the original problem. Finally, we describe a very simple formulation of the solution, whose proof relies on transformations of the plane, a tting ending we think, since a transformation may be regarded as a change in ones point of view. The background developed earlier informs our understanding of this new perspective, and allows us to give a purely geometric description of the transformations needed. For the reader who is unfamiliar with the classical noneuclidean geometries, in which the notions of line and distance are given new interpretations, we provide an overview that is almost entirely self-contained. Such a reader will be introduced to such things as angle excess, stereographic projection, and even a sphere of imaginary radius. For the reader who is familiar with the three classical geometries, we offer some new ways of looking at them, which we are condent will reveal some surprises.

Three problems
Consider three circles in the plane intersecting transversely (that is, with no circles tangent to each other) at a common point, P, as in F IGURE 1. What is the sum of the measures of the angles of triangle of circular arcs ABC? Answer: 180 ! (The picture gives a hint, but we will spell out the solution shortly.) Next consider F IGURE 2, showing three circles whose centers are collinear. Now two (obviously congruent) curvilinear triangles are formed. What can be said about the sum of the angle measures in this case? Answer: This time, the sum is less than 180 ! Finally, consider three circles that intersect in the pattern of a generic Venn diagram, as in F IGURE 3. The boundary of the common intersection of their interiors is a convex curvilinear triangle; the sum of its angles is greater than 180 , because the straightsided triangle with the same vertices lies inside it. What about the other six curvilinear

16
B

MATHEMATICS MAGAZINE

C
P

Figure 1 Three circles through a common point

A C B

B' C' A'

Figure 2 Three circles with collinear centers

triangles formed by these circles? Answer: As the reader might guess, although it is far from obvious from the diagram, the sum in this case is also always greater than 180 !
D E A B C

Figure 3 Three circles in Venn diagram position

The solution to the rst problem admits an elementary proof. Consider triangle ABC in F IGURE 1. Note that by symmetry, the angles labeled with the letter have the same measure, as do those labeled by and by . We then see that 2( + + ) = 360 , hence + + = 180 . The solution to the second problem also admits a simple proof, although a more advanced geometric idea is needed. Namely, the half-plane lying above the line through the three centers may be considered as the upper half-plane model of hyperbolic geometry. In this different sort of geometry, semicircles with centers on the boundary line take the place of lines in Euclidean geometry, and angles between these (hyperbolic) lines are computed using the Euclidean angles between the semicircles. A well-known

VOL. 76, NO. 1, FEBRUARY 2003

17

result in hyperbolic geometry states that the sum of the angle measures of any hyperbolic triangle is always less than 180 . For the third problem, we will also show that the circles are lines in a model of geometry, this time spherical geometry, in which the angle sum of any triangle is more than 180 . More generally, we consider any conguration of three circles that intersect in pairs and give a simple criterion for deciding into which geometry they fall.

The general theorem

Three circles, provided they intersect in the way we have described, determine three lines by their points of pairwise intersection. It may surprise you to learn that these three lines are either concurrent or parallel. The position of the point common to these three lines, as described in the following lemma, is the key to determining which geometry will answer the question about the angle sum. L EMMA . Let c1 , c2 , and c3 be three circles in the plane, with any two of them intersecting in two distinct points. Let l12 , l13 , and l23 be the lines determined by these pairwise intersections. Then the lines {li j } are either concurrent or parallel (in which case we consider them to be concurrent at innity). Furthermore, there are exactly three possibilities for the location of the point of concurrency, P: 1. P lies on all three circles (F IGURE 4a); 2. P lies outside all three circles (F IGURE 4b), possibly at innity; or, 3. P lies inside all three circles (F IGURE 4c). Finally, if P lies outside all three circles, but not at innity, the six tangents from P to the three circles all have the same length. The circle centered at P with this common length as radius is perpendicular to all three of the original circles. If P lies at innity, the line through the centers of all three circles plays this role.

P P P

Figure 4 The three possibilities for the location of point P

Proof. Suppose a line through P intersects a circle c in two points A and B (which need not be distinct). The power of P with respect to c is dened as the signed product (P A)(P B), where P A denotes the directed distance from P to A. We invite the reader to prove, using similar triangles, that the power does not depend on the line chosen. (A proof may be found in Coxeter and Greitzer [2, Theorem 2.11].) The power of P with respect to c is positive if P is outside c, negative if P is inside c, and 0 if P lies on c. Viewing P A and P B as vectors, we can equivalently dene

18

MATHEMATICS MAGAZINE

the power as the scalar product P A P B; since the vectors are parallel, the cosine of the angle between them is 1 according to whether they point in the same or opposite directions. If two circles intersect at points A and B, we deduce (by considering line P A, say) that a point P has the same power with respect to both circles if and only if it lies on line AB. Suppose rst that l12 and l23 are not parallel and let P be their point of intersection. Then P has the same power with respect to all three circles; hence, P lies on l13 and the lines are concurrent. If, on the other hand, two of the lines are parallel, an argument by contradiction shows that all three must be. If the lines intersect, then, as we have just observed, the power of P with respect to all three circles is the same. Denoting this common power by P , we have: Case 1. If P = 0, then P lies on all three circles. Case 2. If P > 0, then P lies outside all three circles. Case 3. If P < 0, then P lies inside all three circles. Finally, if P lies outside, then the length of a tangent from P to any of the circles is P . The circle with this radius and center P is orthogonal to all three circles. As an interesting digression, we note that for a general pair of circles, which need not intersect, a point has the same power with respect to both if and only if it lies on a particular line, called their radical axis, orthogonal to the segment joining their centers. In the course of computing the equation of this line in Cartesian coordinates, Coxeter and Greitzer [2, Section 2.2] also prove the beautiful fact that, if the equation of a circle is put in the standard form F(x, y) = (x a)2 + (y b)2 r 2 = 0, then the power of any point (x, y) with respect to that circle is just F(x, y)! This result is not unexpected, since the power is constant on circles concentric with the given one. We are now prepared to state the theorem. Triangle means a curvilinear triangle that is not subdivided by any arcs of the three circles. In case one, only the triangle that does not include P as a vertex is considered. T HEOREM . Let c1 , c2 , and c3 be three circles in the plane, with each pair intersecting in two distinct points. Let l12 , l13 , and l23 be the lines determined by these pairwise intersections, with common point P. Then the sum of the angles of any triangle formed by the three circles is determined according to the three cases of the lemma: 1. if P lies on all three circles, the sum is equal to 180 ; 2. if P lies outside all three circles, the sum is less than 180 ; and, 3. if P lies inside all three circles, the sum is greater than 180 . The proof of the theorem requires some knowledge of the classical noneuclidean geometries, to which we now turn our attention. The reader who is already familiar with these is invited to skip ahead to the section in which the theorem is proved as follows: For any of the possible congurations, we give a conformal map that takes our three circles to the lines of one of the three classical geometries; since the map is conformal, the angle sum is preserved, and thus found to be equal to, less than, or greater than 180 accordingly.

A quick tour of three geometries

A geometry is an abstract mathematical system in which the undened notions of point and line are assumed to behave in accordance with certain axioms. (Euclidean

VOL. 76, NO. 1, FEBRUARY 2003

19

and non-Euclidean Geometries, by Marvin Greenberg [4], presents a lively account of the historical development of the classical geometries. Edwin Moises text, Elementary Geometry from an Advanced Standpoint [7], is another very comprehensive reference.) In practice, we visualize a geometry by working with a model, which is described more concretely. Objects in the model represent points and lines in such a way that the axioms of the system are fullled. Although there is much to say about the classical geometries, we continue our tale of three circles and focus on the sum of angles in a triangle. The Cartesian plane, consisting of pairs of real numbers, is a model for Euclidean geometry, provided we adopt the usual concept of distance. Each pair of numbers represents a point, while lines are the solution sets of linear equations. Using algebra to study lines, points, and circles is called the analytic method. That the angle measures in a triangle sum to two right angles can be derived either from Euclids axioms or an analytic approach. The Cartesian plane and its physical approximations, such as tabletops, in which parallel lines remain equidistant and meet a common transversal line at the same angle on the same side, have historically been described as at (as opposed to a sphere or other curved surface); thus, Euclidean geometry is said to be at, or have curvature zero. Any surface on which a distance function, or metric, has been dened is a model of some geometry. (These notions may be extended to higher dimensions as well.) The lines on this surface, called geodesics, are the curves that locally realize the shortest distances between their points. Spherical geometry A sphere sitting in R3 , where we know how to measure distances, inherits a metric that measures distances along the surface. Here, the lines are the great circlesthose circles cut by planes through the spheres center. The word local in the denition of geodesic is important; you must choose the short way around. The geodesic nature of great circles can be understood intuitively by noting that the distance along an arc is proportional to the central angle it subtends, and that central angles obey a sort of triangle inequality: where three planes meet, the sum of any two of the face angles is greater than the third. Thus, a path made up of very small (think innitesimal) arcs will be shortest if all the arcs lie in the same plane through the center. A rigorous proof, as suggested by the preceding discussion, requires integration and other concepts from calculus. (Geometry from a Differential Viewpoint by John McCleary [6] is an accessible introduction to the application of calculus to geometry.) Of particular importance to us is the sum of the angles of a spherical (geodesic) triangle. Our calculations will be nicer if we measure angles in radians, which we do from now on. Some experimentation, which we invite the reader to do, suggests that the angle sum of a spherical triangle always exceeds and decreases with the triangles area, approaching, but not reaching, as this area approaches zero (and approaching, but not reaching, 5 as the triangle lls up the whole sphere). This observation suggests that we focus on the amount by which the angle sum of a spherical triangle exceeds , which we call its angle excess. We now prove that a triangles excess is always positive by precisely examining its relationship with the triangles area. Several facts will lead us to the correct relationship. First, notice that the angle excess is additive: if a triangle is subdivided into two smaller triangles, the excesses of the component triangles add up to the excess of the whole, as the reader can calculate. Second, congruent triangles clearly have the same angle excess. These are the essential properties of area: the areas of the parts of a subdivided region add up to the area of the whole, and congruent regions have equal area! Moreover, a sphere, like the plane, is homogeneous: any triangle can be rigidly moved, by rotations and reections of the

20

MATHEMATICS MAGAZINE

sphere, to any other position without changing distances or angles (or area). Together, these facts suggest that, for a spherical triangle, angle excess is a constant multiple of area.

Figure 5 The sector swept out by a spherical angle

To prove that angle excess is proportional to area, observe that the angle between two great circles is proportional to the area of the sector they bound. (Note the essential role of homogeneity here!) On a sphere of radius R, the area of the sector swept out by an angle is (/)(4 R 2 ) = 4R 2 . (See F IGURE 5a. For simplicity, we have used the same symbol to represent both the angle and its measure.) Let represent the area of a triangle with angles , , and . The sectors swept out by , , and cover the sphere redundantly; the triangle and its antipodal image are each covered three times, while the remainder of the sphere is covered exactly once (F IGURE 5b). Thus, with a little algebra, we obtain the formula . R2 In particular, on a sphere of unit radius, the angle excess is exactly equal to the area. + + = Hyperbolic geometry A sphere is said to have constant positive curvature, and it is easy to imagine that a small sphere has large curvature, while a large sphere has small curvature. What would it mean for a surface to have constant negative curvature? This question will lead us to a model where the sum of the angles in a triangle is always less than two right angles. To see how we might describe a surface of negative curvature, we start with some formal manipulations on the equation x 2 + y 2 + z 2 = R 2 . As the constant R 2 increases toward +, the curvature of the sphere described by the equation diminishes toward zero. We may view the Euclidean plane as a sphere of innite radius. What happens when the constant term passes innity and reappears on the negative end of the number line? We will call a surface satisfying an equation of the form x 2 + y 2 + z 2 = R 2 = (i R)2 , which has been aptly described as a sphere of imaginary radius, a pseudosphere. Of course, the equation will have no solutions unless we expand the domain of our variables from the real to the complex numbers, and its radius makes no sense unless we expand our notion of distance to include imaginary numbers. Distance and angles in Cartesian space are measured via the dot product, familiar from multivariable calculus. This product is an example of a symmetric, bilinear form, a type of operation that plays a large role in many branches of mathematics. In addition,

VOL. 76, NO. 1, FEBRUARY 2003

21

the dot product is positive denite: for any vector v, v v 0, and v v = 0 if and only if v = 0. The length of a vector v, denoted |v|, is dened as v v and the angle between two vectors is determined by the formula cos() = v w/(|v||w|). If we allow the coordinates of our vectors to be any complex numbers, the dot product remains a symmetric, bilinear form, although it is no longer positive denite. A form with these properties is called a pseudo-metric. It is this form that we choose to measure distances and angles in complex Cartesian space, Cn , allowing these quantities to have complex values. Returning to the equation x 2 + y 2 + z 2 = R 2 , we consider, to obtain a twodimensional surface, only those solutions for which x and y are real. The remaining variable, z, is then forced to be purely imaginary, so we let z = it. To picture the pseudosphere, we map the solution set of our equation to a surface in R3 , using the map (x, y, it) (x, y, t). This image is a two-sheeted hyperboloid, the solution set of the equation x 2 + y 2 t 2 = R 2 . The sheets are analogous to the two hemispheres of an ordinary sphere, the origin (0, 0, 0) may be thought of as the center, and the points (0, 0, R) as the North and South poles. In order to measure distances and angles, we must remember that our space is an image of a subspace of C3 , where the actual distances and angles are measured using the dot product. Since a vector (x, y, t) actually represents the vector (x, y, it), its actual length is x 2 + y 2 + (it)2 = x 2 + y 2 t 2 . Similarly, the angle between two vectors (x1 , y1 , t1 ) and (x2 , y2 , t2 ) must be based on the form x1 x2 + y1 y2 t1 t2 . Some readers may recognize this as a Lorentz metric, the three-dimensional version of the pseudo-metric of relativistic space-time. (For a readable and physically motivated, but advanced, introduction to pseudo-metrics, see Semi-Riemannian Geometry, by Barret ONeill [8].) The apparent distances and angles in our picture are distorted from their actual values; for example, the true length of every radial vector from the origin to the surface is the imaginary number i R! To measure distances and angles on the pseudosphere itself, we apply the pseudometric to its tangent vectors. For any point P in R3 , the tangent space at P is the copy of R3 consisting of all vectors emanating from P. The tangent line at P to a curve through P on the surface is a one-dimensional subspace of this tangent space. The tangent plane to a surface at P is the two-dimensional space composed of all these lines. It is useful (and intuitive) to write coordinates and other quantities related to tangent vectors in terms of differential expressions. If f is a differentiable, real-valued function on R3 , students know how to compute f , and use it to nd directional derivatives by taking dot products. We prefer another vocabulary: the differential of f is the function that takes a tangent vector v P , at a point P, to the real number f (P) v P . This gives the best linear approximation to the change in the value of f that results from starting at P and travelling with a displacement v P . In particular, the Cartesian coordinates, x, y, and t in our model may be viewed as functions of P R3 , and dx, dy, and dt, are the differentials of these functions. If we suppress the vector argument of these differentials, we can use dx, dy, and dt as the rst, second, and third coordinates of a general tangent vector with respect to a parallel coordinate system based at P. (In F IGURE 6a, the linear approximations dx, dy, and dt happen to be exact, since orthogonal projection onto an axis is a linear function.) Thus we may conveniently write the (Lorentz) length of a tangent vector as the differential expression dx 2 + dy 2 dt 2 . This gives us the metric we need in order to talk about the lines of this geometry, which are the geodesics of this metric. A beautiful fact is that this differential expression is real and positive, when applied to vectors tangent to the pseudosphere. Since the length of a path is computed by

22

MATHEMATICS MAGAZINE

integrating the lengths of its tangent vectors, paths between points on the pseudosphere have real, positive length, and it makes sense to talk about geodesics as paths of locally minimal length. To see that dx 2 + dy 2 dt 2 > 0, it helps to use cylindrical coordinates. As an orthonormal basis for the tangent space to R3 at the point with cylindrical coordinates (r, , t), take a radial unit vector, a circumferential unit vector in the counter-clockwise direction, and a vertical unit vector; with respect to this basis, the coordinates of a tangent vector are (dr, r d, dt). (The factor of r in the circumferential coordinate results from the fact that changing the central angle by a small amount changes the distance travelled by r times that amount. See F IGURE 6b.) The expression dx 2 + dy 2 , the squared length of the vectors horizontal component, is equal to dr 2 + r 2 d 2 . For a vector tangent to the hyperboloid, dt 2 < dr 2 , since the radial slope of the hyperboloid is less than that of the cone to which it is asymptotic. (Algebraically, it follows from the equation r 2 t 2 = R 2 that r 2 < t 2 , and by differentiating we calculate that dt 2 = (r 2 /t 2 ) dr 2 .) Thus dx 2 + dy 2 dt 2 = r 2 d 2 + (dr 2 dt 2 ) > 0.

P x y t (0,0,0)
a

rd
d

(0,0,t) t (0,0,0)
b

Figure 6 Comparison of two bases for the tangent space at P

The geodesics turn out to be the intersections of the pseudosphere with planes through the origin, analogous to those on an ordinary sphere, although it is harder to see why this is so, and we wont prove it here. The pseudosphere is also homogeneous, and the angle sum of a triangle satises a completely analogous formula: + + = R 2 .

It follows that this sum is always less than . Of course, there is a different pseudosphere for each value of R. By analogy with the sphere, one might guess that a small value of R yields a pseudosphere of large negative curvature, while a large value of R gives a pseudosphere that is so little curved as to be nearly at. However great or small the curvature, the sum of the angles in a triangle is still less than two right angles. (A book by B. A. Dubrovin, A. T. Fomenko, and S. P. Novikov [3] gives a thorough discussion of both the sphere and pseudosphere.) Each sheet of the pseudosphere is a model of a geometric object called a hyperbolic plane. It would be nice to be able to see this plane looking more like a plane, without having to work with an object as complicated as the Lorentz metric. There is a planar map of the pseudosphere that shows angles accurately, obtained by a method called stereographic projection. A similar map is also available for the ordinary sphere. Any

VOL. 76, NO. 1, FEBRUARY 2003

23

map that preserves angle measure is called conformal. Since our tale of three circles involves angle measure, conformal maps will be a powerful tool.

Stereographic projection
The sphere To obtain a conformal mapping of the ordinary sphere onto the plane, project from any point of the sphere onto the plane tangent to its antipodal point. Any such map, or its inverse map from the plane to the sphere, is called a stereographic projection. Every circle on the sphere (not just the great circles) projects stereographically to a circle or line (which may be thought of as a circle through ) in the plane, and it maps to a line if and only if it passes through the point of projection (which maps to ). Conversely, every line or circle in the plane is the image of a circle on the sphere. (See F IGURE 7, and note there that the center of a circle on the plane is not, in general, the projection of the center of the corresponding circle on the sphere, but rather the apex of the cone tangent to it.) These valuable properties may be proven by elementary arguments. (The arguments are outlined and nicely illustrated in Hilbert and Cohn-Vossens classic book, Geometry and the Imagination [5, pp. 248251].)
N

Figure 7 Stereographic projection preserves angles and takes circles to circles or lines

To see that stereographic projection from the sphere to the plane is conformal, consider F IGURE 8a, which shows the cross-section of the sphere cut off by a plane through the point of projection, N , its antipodal point, S, and another point P on the sphere; (P) is the projected image of P, and A is the point where the cross-section of the plane tangent to the sphere at P intersects the tangent plane at S. An angle with vertex at P in the plane tangent to the sphere is cut by two planes intersecting along line N P. Since (P)P A = P(P)A, these planes cut the same angle at (P) in the horizontal plane through S, by symmetry. To see this, imagine an angle formed by two stiff planes of paper; if you snip at the same angle to the spine in either direction, the angle you make is the same. (The projected image of the angle is its reection in the plane through A that is perpendicular to N P(P).) F IGURE 8b illustrates the local behavior of stereographic projection. If Q is a nearby point on the tangent plane through P, then N Q P is approximately similar to N (P)(Q). Therefore, stereographic projection is linearly approximated at each point P by a dilation (uniform scaling). The dilation factor varies with the latitude of P, increasing without bound as P approaches N , with a minimum value of 1 at S.

24
N
N

MATHEMATICS MAGAZINE

P (Q)

S
a

(P)
b

(P)

Figure 8 Stereographic projection is conformal and locally approximated by a dilation

The pseudosphere To obtain a conformal map of the pseudosphere onto the plane, project each point of the hyperboloid model onto the horizontal plane through the South pole, (0, 0, R) (that is, the plane t = R), via the line connecting it to the North pole, (0, 0, R), as in F IGURE 9. This projection maps the southern hemisphere onto the interior of the disk of radius 2R centered at the origin, while the northern hemisphere, except for the North pole, goes onto the disks exterior. The disks interior is known as a Poincar Disk model of a hyperbolic plane, and its boundary is called the e circle at innity. It can be shown that each geodesic maps to a circle or line orthogonal to the circle at innity (with the points of intersection removed); the geodesics through the poles go to lines. (See B. A. Dubrovin, A. T. Fomenko, and S. P. Novikov [3] for a proof.)

Circle at infinity

Poincare Disk Model

!

Figure 9 Projection of the hyperboloid model of the pseudosphere onto a plane

The conformality of the projection of the hyperboloid model cannot be demonstrated by elementary geometric arguments because the angles on the hyperboloid are measured using a different bilinear form than the one used to measure angles in the plane. So to prove that angle measure is preserved, we must resort to calculation to compare the angles between tangent vectors to curves, before and after projection. The reader who wishes to just believe us and avoid the technicalities involved may skip the

VOL. 76, NO. 1, FEBRUARY 2003

25

calculation below without any loss of continuity. For those who wish to venture in, the proof provides a nice application of calculus techniques to geometry. Proof that projection to the Poincar disk model is conformal We restrict our e attention to the southern hemisphere; the proof for the northern hemisphere is similar. Let denote the projection map, extended to the region t < R, and viewed as a map onto R2 by ignoring the last coordinate (R) in the image. Suppose curves and intersect at P = (x, y, t). The derivative of at P, D(P), carries vectors tangent to and at P to vectors tangent to their projected images at (P) (by the chain rule). Let v P , w P L denote the Lorentz product of tangent vectors v P and w P at P, that is, (x1 , y1 , t1 ), (x2 , y2 , t2 )
L

= x1 x2 + y1 y2 t1 t2 .

We sketch a proof that, if v P and w P are tangent to the hyperboloid, then D(P)(v P ) D(P)(w P ) = [4R 2 /(R t)2 ] v P , w P L . In other words, the dot product of the images is just scaled by a constant factor from the Lorentz product of the preimages. It follows that the angle, , between v P and w P , which is determined by cos = vP , wP vP , vP
1 2

L
1 2

wP , wP

is equal to the angle, , between D(P)(v P ) and D(P)(w P ), which is determined by cos = D(P)(v P ) D(P)(w P )) [D(P)(v P ) D(P)(v P )] 2 [D(P)(w P ) D(P)(w P )] 2
1 1

since the scaling factor cancels out. In contrast to the sphere, the scaling factor decreases as P moves away from S (t decreases), with a maximum value of 1 at S.
N=(0,0,R)

2R (P) t=-R R-t

P=(x,y,t) r

Figure 10 Effect of the projection on the radial coordinate

Referring to similar triangles, as in F IGURE 10, we see that the image under of the point P with cylindrical coordinates (r, , t) is the point whose polar coordinates 2Rr are ( Rt , ). The derivative at P of the conversion from Cartesian to cylindrical coordinates takes (dr, r d, dt) to (dr, d, dt). If : R3 R2 is the map (r, , t) 2Rr ( Rt , ), its derivative (or Jacobian, if you prefer) at (r, , t) is D(r, , t) =
2R Rt

0 1

2Rr (Rt)2

26

MATHEMATICS MAGAZINE

Applying D(r, , t) to (dr, d, dt), followed by the derivative at (r, , t) = 2Rr ( Rt , ) of the conversion from polar back to Cartesian coordinates, we calcu2R late (thanks to the ever-valuable chain rule) that D(P)(dr, r d, dt) = Rt (dr + r dt , r d). Rt For any point P on the hyperboloid, r 2 t 2 = R 2 , and for any vector tangent to the hyperboloid at P, dt = (r/t) dr . After some simplication using these 2R substitutions, we nd that D(P)(dr, r d, (r/t) dr ) = Rt ((R/t)dr, r d). The reader can now check that for two tangent vectors at P, D(P)(v P ) D(P)(w P ) = [4R 2 /(R t)2 ] v P , w P L , as stated. Sequences of stereographic projections Stereographic projection is an indispensable tool for transforming geometric models without changing the angles between geodesics. In particular, it provides us with another celebrated model of the hyperbolic plane called the Poincar half-plane. To obtain it, rst project the Poincar disk of rae e dius two from a horizontal plane onto the southern hemisphere of the unit sphere (and its complement, including the point at , onto the other hemisphere). Then project onto any vertical plane tangent to the equator. The image of the circle at innity under this sequence of projections is called the line at innity. Each half-plane is an image of a hyperbolic plane, as in F IGURE 11.

Figure 11 Conformal transformations between the Poincar disk, hemisphere, and e Poincar half-plane models of hyperbolic geometry e

By a sequence of two stereographic projections from different points, we also obtain a conformal map of the Euclidean plane in which the images of the geodesics are either lines or circles. See F IGURE 12. In summary, we now have maps of the plane, sphere, and pseudosphere in which the geodesics are represented by lines and circles. Just as a map of the earth must be distorted in order to print it on the at pages of an atlas, our maps distort the true distances between points in the geometric objects they depict. The art of map-making revolves around choosing a projection whose particular type of distortion allows the map to be useful for its intended purpose. For example, the famous Mercator projection of the earths surface is useful to navigators because it accurately depicts the compass

VOL. 76, NO. 1, FEBRUARY 2003

27
N

Figure 12 Euclidean geometry on the sphere

bearing between any two points. (McCleary [6, chapter 8bis ] gives a nice discussion of map projections.) Since we are interested in the properties of angle measure, we have chosen to view our circles through conformal maps, which render the true angles between smooth curves. The question remains: Given a trio of circles that intersect in pairs, how can we interpret the conguration as a geodesics on a map of one of the geometries we have studied? Is such an interpretation even possible?

Choosing a geometry: a proof of the theorem

Recall that P is the intersection of the three lines determined by the pairwise intersections of the three circles. We will show that, depending on the location of P relative to the circles, there is a conformal map that takes the circles to lines of a standard geometric modela model for which we know a great deal about angle sums. Case 1: P lies on all three circles. In this case consider a sphere tangent to the plane with South pole at P, and a second plane tangent to this sphere at the North pole, as in F IGURE 12. The images of our three circles on the sphere are circles through the South pole. If we project again, this time from the South pole of the sphere onto the second plane, these circles are taken to straight lines. Thus, c1 , c2 , and c3 are geodesics in a conformal planar model of Euclidean geometry, and the sum of the angles of the triangle they form is 180 . Case 2: P lies outside all three circles. We already studied the special case where the three lines determined by pairwise intersections of the circles are parallel; this happens when the centers of all three circles are collinear, and that line was seen to be the boundary line of a hyperbolic plane. Having dispensed with that case, we assume that the lines are concurrent at a point P lying outside of all three circles. In this case, the Lemma presents us with a circle, d, that is orthogonal to all three circles. Thus, each arc of the circles lying inside or outside d is a geodesic in a model of a hyperbolic plane, and thus the sum of the angles of any triangle they form is less than 180 . Case 3: P lies inside all three circles. Consider the family of spheres (of varying size) tangent to the plane with South pole at P. Consider how the area of the stereo-

28

MATHEMATICS MAGAZINE

graphic projection of the disk bounded by c1 onto each of these spheres compares to the surface area of the sphere: if the sphere is very small, the projected region will have area more than half that of the sphere (in fact, it will include the entire Southern hemisphere); if the sphere is very large, the projected area will be less than half the surface. This is just a matter of having c1 lie inside or outside the preimage of the equator on the plane. (See F IGURE 13.) Allowing the sphere to vary continuously, we see that there is a unique sphere, S , such that the projected area on S is exactly half the surface area. (For an alternative argument, see the third remark below.) Consequently, the image on S of c1 is a great circle.

N P
c1
N

c1

Figure 13 The family of spheres tangent to the plane at P

We claim that the images on S of c2 and c3 are also great circles. It sufces to prove this for c2 , as the same argument works for c3 . To do so, observe that the image of l12 under stereographic projection is a meridian, that is, a great circle passing through the North and South poles of the sphere. Since the image of c1 is also a great circle, the images of the points of intersection of c1 with l12 are antipodal. Since the points of c1 l12 also lie on c2 , it follows that the image of c2 is a great circle. F IGURE 14 illustrates this.

c1 c2 l12

Figure 14 The images of the circles on the sphere S are great circles

VOL. 76, NO. 1, FEBRUARY 2003

29

We have thus shown that the images on S of the three circles are great circles; that is, they are spherical geodesics. Thus c1 , c2 , and c3 are geodesics in a conformal planar model of spherical geometry, and the sum of the angles of any triangle formed by them is greater than 180 . R EMARK . That Euclidean geometry occurs only in the instance that point P lies exactly on the circles is an illustration of the fact that our familiar at geometry is just a single point in the spectrum of geometries, from the sphere of large radius R, to the at plane, where R is effectively innite, to the pseudosphere, whose radius is imaginary in the model we have shown. R EMARK . Using the sequence of stereographic projections described earlier, we may transform the general picture for Case 2 into the half-plane picture. continuity argument in Case 3 of the Theorem, it R EMARK . Although we like the may be avoided as follows. Let s = P (P). The sphere S in the above proof is the one that has radius s/2. To show this, consider, for each of the circles c1 , c2 , and c3 , the chord through P that is perpendicular to the radius through P. Each of these chords has length 2s and midpoint P. Let e be the circle centered at P with radius s. This circle projects stereographically to the equator of S , and c1 , c2 , and c3 , which intersect e at diametrically opposite points, also project to great circles. This alternative argument emphasizes the parallels between Cases 2 and 3. R EMARK . The theorem remains true if circle is understood in its more general sense to include straight lines as well. We invite the reader to extend the theorem to the limiting cases in which two or more of the circles are tangent. The possibilities are illustrated in F IGURE 15. Case I

IIa

IIb

IIc

Figure 15 The possibilities for two circles to be tangent

30

MATHEMATICS MAGAZINE

of the gure leads to Euclidean geometry, when viewed with the right map: One vertex of the triangle is thrown out to innity by the map that takes the circles to straight lines; hence, two sides of the triangle become parallel lines. The possibilities in Case II are hyperbolic, with one or more vertices of the triangle lying on the circle (or line) at innity. What is the measure of an angle whose vertex lies on the circle at innity? Note that none of the limiting cases is spherical.

Changing your point of view: a transformational approach

We are indebted to Keith Burns for pointing out the following very elegant formulation and proof of the Theorem. His proof uses the group of M bius transformations of the o extended complex plane (that is, the plane, regarded as the eld of complex numbers, together with a point at ). M bius transformations are invertible, bicontinuous, and o conformal, and take generalized circles (with lines regarded as circles through ) to circles. Moreover, given any two ordered sets consisting of three points each, there is a (unique) M bius transformation taking each point of one set to the corresponding o point of the other. (The excellent, very readable text on geometry from a Kleinian point of view by Brannan, Esplen, and Gray contains a thorough discussion of the M bius o group and its geometric properties [1, Chapter 5].) In particular, and this is all we will need, there is a M bius transformation taking o any given point to . As we have seen, a transformation with the required properties can be constructed by composing a pair of stereographic projections. Consider a curvilinear triangle ABC formed by the three circles c1 , c2 , and c3 . Without loss of generality, assume that A lies on c1 and c2 , and let A be the other point of intersection of c1 and c2 . We then have three possibilities for the positions of A and A with respect to the third circle, c3 , which correspond to the three cases of the Lemma and Theorem: A lies on c3 ; A lies on the opposite side of c3 from A; or A lies on the same side of c3 as A. By simply changing our point of view, placing A at , we can discern by inspection how the angle sum of triangle ABC compares to 180 . Although this approach does not demonstrate the underlying global geometry in which the circles are geodesics, it does have the advantage of being delightfully simple and direct. T HEOREM . (A LTERNATIVE F ORMULATION ) Let c1 , c2 , and c3 be three circles in the plane, with each pair intersecting in two distinct points. Then exactly one of the following three conditions holds and determines the sum of the angles of any triangle formed by the three circles: 1. the intersection of each pair of circles contains a point of the third circle, in which case the sum of the angles is 180 ; 2. the intersection of each pair of circles lies entirely inside or outside the third circle, in which case the sum of the angles is less than 180 ; or, 3. the intersection of each pair of circles contains one point inside and one point outside the third circle, in which case the sum of the angles is greater than 180 . Proof. Let A, B, and C be the vertices of a triangle formed by c1 , c2 , and c3 , with {A, A } = c1 c2 , as in the paragraph preceding the statement of the theorem. The angle sum of the triangle does not depend on which pair of circles we consider, so it sufces to show that this sum is determined by the positions of A and A relative to c3 . Apply a M bius transformation, , that takes A to . Under this transformation, o the images of c1 and c2 , which pass through A , are lines. If A lies on c3 , then the image of c3 is also a line, hence the angle sum of the image of triangle ABC is 180 .

VOL. 76, NO. 1, FEBRUARY 2003

31

B A C
() (C)

()

A'

Case 1

A B C

() (C)

()

A'

Case 2

B A C

()

()

(C) A'

Case 3

Figure 16 The effect of a Mobius transformation, , taking A to

If A and A lie on the same side of c3 , then the image of c3 is a circle with the image of A outside it, since is outside; hence the angle sum of the image of triangle ABC is less than 180 . If A and A lie on opposite sides of c3 , then the image of c3 is a circle with the image of A inside it; hence the angle sum of the image of triangle ABC is greater than 180 . The three possibilities are illustrated in F IGURE 16.

Conclusion
Each of the classical geometries, Euclidean, spherical, and hyperbolic, has a variety of conformal representations on the plane, obtained by stereographic projection. Rec-

32

MATHEMATICS MAGAZINE

ognizing these representations enabled us to classify the generalized triangles whose sides are either segments or circular arcs as belonging to one of three types of geometries, allowing us to make the correct conclusion about the sum of the angles. Finally, by using a group of conformal transformations of the extended plane, we were able to rene our solution to be very simple, although perhaps less informative. Our success is just one example of the value of studying noneuclidean geometries and transformation groups.
Acknowledgments. In addition to Keith Burns, we would like to thank Duane Broline, Leo Comerford, Bob Foote, Yuri Ionin, and Rosemary Schmalz for their helpful comments. We would also like to thank an early referee for pointing out the argument in the third remark.

REFERENCES
1. D. Brannan, M. Esplen, and J. Gray, Geometry, Cambridge University Press, Cambridge, 1999. 2. H. S. M. Coxeter and S. L. Greitzer, Geometry Revisited, Random House, New York, 1967. 3. B. A. Dubrovin, A. T. Fomenko, and S. P. Novikov, Modern GeometryMethods and Applications. Part I. The geometry of surfaces, transformation groups, and elds, Graduate Texts in Mathematics # 93, Springer-Verlag, New York, 1992. 4. M. J. Greenberg, Euclidean and non-Euclidean Geometries, Freeman, New York, 1993. 5. D. Hilbert and S. Cohn-Vossen, Geometry and the Imagination, Chelsea, New York, 1952. 6. J. McCleary, Geometry from a Differentiable Viewpoint, Cambridge University Press, Cambridge, 1994. 7. E. Moise, Elementary Geometry from an Advanced Standpoint, Addison-Wesley, Reading, 1990. 8. B. ONeill, Semi-Riemannian Geometry, with Applications to Relativity, Academic Press, New York, 1983.

Pete, Repete, and One Bagel

Pete: Repete: Pete: Repete: Pete: Repete: Pete: Repete: Pete: Repete: Do you want a bagel? Oh, no, I couldnt eat that much. I could halve the bagel. Yes, you should have it. No, if we halve the bagel, we could each have half a bagel. Oh, OK. But that brings up a tricky question. Yes? Before you halve a bagel, a bagel will have a hole. After you halve the bagel, does half a bagel have half a bagel hole, or it is a whole hole? A whole! But if half a bagel is to have a whole hole, then when you halve a whole bagel, you dont, in fact, halve a bagel hole, since in each half you have a whole hole, which is two holes. Mysterious! So when you halve a bagel, you get to have a bagel half and a bagel hole. Right! This means that when you have a whole bagel, you get half the bagel holes you get when you halve a whole bagel! Holy Cow! G LENN A PPLEBY S ANTA C LARA U NIVERSITY S ANTA C LARA , CA 95053

Pete: Repete: Pete: