with common corresponding sides and vertices is called a develop- ment of the original polyhedron R. If we glue these polygons together along their comnion sides, we recover the polyhedron, By construction and definition we have the following. “ The development of a convex polyhedron is a system of polygons satisfying the following conditions: petem of pos ()) Each side is common to exactly two polygons; each vertex is common to at least three polygons. (2) The numbers n of polygons, m of distinct vertices, and | of dis- tinct sides satisfy the Euler relat " ” men—1=2. (3) Any two polygons Q and Q may be connected by some sequence of polygons 0 = Qo, Ox,..-, Qs = Q' for which Qo and Qs, Oy and Qe, Qi-1 and Q; have common sides. oe (4) Common sides of two polygons have equal lengths. (S) The sum of the angles at a common vertex of several poly s ral polygons is less than 2x (because the sum of the plane angles of a convex polyhedral angle is less than 27), Inhis monograph Convex Polyhedra, A. D. Aleksandrov proved the following remarkable converse theorem: Auexsaspnov’s misonem, A system of polygons satin condi fon (43 abort a develope of ome cane ob heen Notice the connections between this theorem and Cauchy's theorem of section 20. Aleksandrov's theorem proves that there exists 1 convex polyhedron with a given development, and Cauchy's theorem proves that the polyhedron is unique (up to congruence and symmetry). 4, Linear Systems of Convex Figures 25. LINEAR OPERATIONS ON POINTS In this chapter we assume that the reader is familiar with ele- mentary analytic geometry and analysis ‘Linear operations may be performed on vectors; that is, vectors may be added to vectors, and vectors may be multiplied by real. mumn- bers (scalars). We shall show how these operations may also be per- formed on convex solids. Let ns consider each point q of the plane or of space as the end point of a vector issuing from some fixed point ©, called the origin. The letters ©, 9, 8 will denote figures, 0, 4, B, C,p, q,r will denote points; / will denote a line; and s, x, y will denote real numbers. Also, subscripts and primes will be used when needed. Ifa vector OF is the sum of vectors Og and Og, that is, Or = 09 + Op. then we shall call its end point, r, the sum of points q and 4 Gig. 104), and write: raq4+4 Fig. 104 imber, is defined ‘The product s9, where q is a point and s is a res analogously. Namely, r = s9.is the end point of the vector OF = 509,From these definitions, we have: The tine segment connecting two points q and qx (Fig, 105) is th set of all points q, defined by the formula _— q=99+(1-5)q OSS< 1), a or, with s = 1 —s, Ha 9945 (S+51= 155 51> 0), a Fig. 105 Point q, divides segment gq; in the ratio (1 — s)/s or s/s. If $= 51 = 4, then q, = gy is the midpoint of segment 9g. Letr, r1, and rs be the projections of points ¢. :, and q, on some straight line or plane (Fig. 105). Then point r, lies in segment rry and divides iin the seme ratio as divides gg, namely (1 — 5) lence, 2) If points q and gs (Fig. 106) have the coordinates (x, y) and (x, 91) in the plane, then point g, on segment gq: has the coordi- nates (x. Ye) given by Mes sx + Str yea sy + sy This assertion is a consequence of equation (2). If we project seg- ment gg; on the coordinate axes Ox and Qy, then each point of the segment is projected onto the points corresponding to its abscissa and ordinate. Suppose that we are given some function /. Its graph is the graph of the equation |} += lands, 51> 0) @ =f: Deriimion. A function is called a concave function if its graph either lies above or coincides with any chord AB connecting two of its points (Fig. 107). In Figuge 107, let x and x1 be the abscissas, y = f(x) and y1 = fa) be the ordinates of points A and B. Each abscissa between x and x, may be given in the form Masrd sam (6+ 515551 > 0). @ Point C on chord 4B with abscissa x; has ordinate Fe= 9 + sy = sf) + if). Point C; on the graph of y = f(x) with the same abscissa x has ordinate f(x). Since point C; cither lies above C or coincides with it, its ordinate is greater than or equal to the ordinate of C: SO) > sf + sf). o Hence, we could give the following equivalent definition of th concavity of f: A function fis concave if inequality (5) holds for al pairs of numbers x and x1, and for all x, between them given in equation (4).26. LINEAR OPERATIONS ON FIGURES; “MIXING” FIGURES DEFINITION. Suppose that we are given two figures (or two solids) Mand 8. Then their sum, + ®, will be the set of all points of the form p + q where p belongs to and q belongs to ®, For example, if Wis a segment of the x-axis with length 1 and With the origin as its left end point, and is a segment of the y-axis with length 1 and with the origin as its lower end point, then & +B is a square region with sides % and 8. Also, if 8 consists of only one point go, then % + % = W 4 qo is the set of all points p + qs, where p is any point in 8, Notice that & + qy is obtained by trans, lating 4 by the vector gp We can also define the operation of multiplication by a real umber 5 Derivini0N. The product figure, 5%, is the set of all points sp where p belongs to %. Figure oY is obtained from & by a similarity transformation with coefficient s and center at the origin In general, if we are given figures th, Ws, %, and & real num- bers s1. 59, -.+5 5%, then Sith + salle + + + S18 is the set of all points Spt spat ++ + supe where p: belongs to %, We shall study only the case in which we are given two figures Q and Q, and two positive numbers s and s; with sum s + s, = 1 Then we write Oy = sD 49M = D+ ~ 50, Thus, Q, is the set of all points of the form g, = sq + sigs, where a belongs to © and q belongs to ;. Geometrically, O, is the locus of all points 4, dividing the segments qq; in the ratio s/s, Derinrtion. The operation of forming ©, as shown above is called mixing © and Q, Ifs = sy = 4, then O, = Q, is the locus of the midpoints of seg- ments connecting ail points of & with all points of Examete 1. Let and f; be parallel lines (Fig. 108) given by the normal equations xoosa+ysina=h, o xeosa+y sina = hy 2 Then hast (645 = 155520) is a line parallel to f and f given by the equation 008 a + sin a = hy @ where hig = 9h + shy Proof, Let q, be an arbitrary point on line ,. Then wag tn where gis on land qs is on Let (x,y), (x1, 71), nd (tn 94) be the co- ordinates of points g, gi,-and q, respectively. Then from (3) of the preceding section, we have bes ox ta wey t synThe coordinates (x, ») of point g satisfy equation ( eos a +y sina = h, Analogously, cos a + yi sina = Consequently, Xe COS + Ye a= (sx + sims) cosa + (sy + sy) sina = s(x cosa + sin a) + si(x1 cos @ +»: sin a) sh + Sihy = hes that is, the coordinates of point g, satisfy equation (3). Hence, if we connect all points q on the line / with all points gx on the line / by segments gg;, and on each of these segments, pick out the point ge which divides the segment in the ratio s1/s, then the set of all points q, is the line /, (Fig. 108). Remark. Now suppose that the plane figures © and Q, do not lie in one plane, but in two parallel planes R and Ry, Let every point q of figure © be connected with every point q: of &; by line segments 4qqx- The set of all these segments forms some solid T; contained be- tween planes R and Ry. Now construct plane Ry parallel to R and Ry, and dividing the distance between them in the ratio s;/s. The intersection of this plane and T gives us 8, = sO + 5,04. In fact, the point of intersection of the plane R, and each segment 9g, divides this segment in the ratio s1/s. Obviously, the entire plane Ry may be defined as a linear combination of planes R and Ry R= sR +R ExaMPLe 2, Now we shall consider the lines / and h; to be not only parallel, but also to have the same direction (Fig. 108). Each of the ines , 1, and /, divides the plane into two half-planes. / divides the plane into half-plane , lying on the right, and half-plane B, lying on the left of /. Analogously, (; divides the plane into half-planes A, and By, lying on the right and left of /s, and /, divides the plane into half-planes 4, and B,, lying on the right and left of J,. We can show that A= A + 5A, a By = 5B + By oO Proof, Half-plane A is the set of all points whose coordinates (x, )) satisfy the inequalit x cosa + y sina > h, © Half-planes A; and A, are the sets of points (x1, 1) and (x, y2) whose coordinates satisfy the inequalities x cosa + yr sina > hy, o 24 60S & + yy Sin a > hy Linear combinations p, = sp + sip: for which the coordinates of points p:(x, y) and: p(x, y) satisfy inequalities (6) and (7) yield points py:(xs, )s), Whose coordinates satisfy inequality (8). Its = s; = 4, then line J is the locus of the midpoints of segments connecting points of lines / and f,. Half-plane 43 lying on the right of Jy is composed of the midpoints of all segments con- necting points of half-planes 4 and A. Thisis apparent geometrically. ments by k and ky, and form the fi kya sk+ sky G+ = 150120) o re Figure fy lies on the line /, = s! + sih. As shown in Example 1, / is parallel to / and fy, The equations of lines f, and f, are xeosa+ysina=h, x cosa + ysin a = hy, where hy = sh + syhy xeosatysina = hy Figure k, is the locus of points gy dividing segments 99: in the ratio 51/3, where q is on k and q, is on ky. As shown in Figure 109, the points g, form segment 4,B, of line J, where A, is the point divid- ing segment 4s in the.ratio s,/ and B, divides BB in the same ratio. If a, a1, and a, denote the lengths of segments k, ks, and ke, thenEquations (10) and (11) show that the relationship between the lengths a, ay, and ds of segments k, ky, and k, and between the dis- tances f, hy, and Ay from the origin to the lines on which these seg- ments lie is the same as the relationship (9) between the segments themselves. ‘We can now prove the following fundament theorem. ‘Tazorem 1. 1f and Qy are convex figures, then S.= +n +0555 20) is also a convex figure. Proof. Suppose that p, and q, are two points of O, (Fig Fig. Then they may be represented in the form Pe=Pt—IPy wag —9q, (2) where p and g are points of figure ©, and p, and q, are points of figure By. Segment p.gg is the set of all points of the form nage thy +H = 14H) ay From equations (12) and (13) we obtain q+ sag) + ta(sp + s3ps) = 51g + Gp) + siliqa + 4p) = 9r ar where r and r, are given by rage tp, Thus ris on segment pq. Since 8s convex, it contains this segment, and must also contain r. Similarly, ry belongs to Oy We have shown that ris of the form sr + syri where r belongs to Aland r; to Or. But O. is by definition the set of all points hav- ing this form. Hence, r, belongs to ©. Since point r, is an arbitrary point of the segment p,q. then the entire sezment peg, belongs 10 £2, Therefore 6, is convex and the theorem is proved THEOREM 2. Let © and £4 be plane convex figures and let ! and 1h, be parallel supporting lines for them having the same direction Form the figure Dy = 30 + s:Qx (5 + 51 = 15 5, 51 > 0) and the line l, = s!-+ Syl. Then ly isa supporting line for 8. parallel to 1 and I and having the same direction. Fig. 111 Proof. Suppose that lies to the left of ! and ©, lies to the left of f: (Fig. 111), From Example 1, leis a line parallel to / and f and having the same direction. From Example 2, we conclude that 8. lies to the left of le Line / and figure © must have some point x in common, be- cause /is a supporting line for 0. For the same reason, /; and Oy have a common point x. Now consider the point x, = sx It belongs to both f, and ©,. Consequently, /, is a supportis for ©, and the theorem is proved. Taronsn 3. If figures © and Q, are translated by vectors Oa and Oa, respectively, then By = sO + 5\Q, is translated by the vector Ois, where a, = sa+ srar. Proof. 0 is transformed into © + a, O; into Q, + a. Then 8, = sD 4 5:61 is transformed into SCD + 0) + 5:(Qa + a) = (9D + Or) + (50 + Span) =D, + ay, and the theorem is proved27. LINEAR SYSTEMS OF CONVEX POLYGONS; ‘AREAS AND “MIXED AREAS” Consider two polygons @ and Q1 in which corresponding sides are parallel and have the same directions (for example, polygons Aida. AsAigs + Ay and Ay'Ag’... AVA’ on. Ait Figure 112). We shall construct polygon Q, = sO + 5.0; where 5 45: = 1 and s,s > Oas usual. Let a; be the length of side 4,4;.1 of the first polygon and a’ that of side 4\'4'c,1 of the second. Form the linear combinations ai) = say + sya for i= 1, 2,..., n. By Example 3, section 26, aj? is the length of where = shi + Al, = Aug + 5A’ Let /and /’ be the supporting lines for Q and Q; which contain seg- ments a; and ay’ respectively. Then /, = sf + sz!’ is a supporting line for Q,. Since segment 4,A;, belongs to both Q, and its support ing line /, then itis part of the boundary of Q,. The boundary of Q, 106 consists of the segments AiAy,1, parallel to and having the same direction as the corresponding sides of Q and Qs, Thus Q, is a polygon with vertices 4," (i = 1, 2,...,n) and sides Ai®A,,1 with lengths ai. If hy hy’, and hy are the perpendicular distances from the origin Oto sides a,, a’, and ai of polygons Q, Qi, and Q,, then o + ih (see Example 3, section 26) We drew Figure 112 wi origin inside both Q and Qs. Ifit were not, we could make it so by translating Q and Qy. As shown in Theorem 3, section 26, Q, would also be translated. Polygon Q, can be constructed as follows: First, connect vertices Ay and Ay’ of polygons Q and Q with segment Ayy’ and divide this segment in the ratio s;/s, to obtain 41, From 4," draw a paraliel to 4,43. When sufficiently extended, it will meet segment ApAq’ at the point Aa!, a vertex of Q,, From Ay! draw a line par- allel to Agoda and extend it to meet segment 434s’ in As. Continue in this manner to construct the remaining vertices and sides of O, the last to be constructed will be the side 4,04 ‘Now suppose that polygons Q and Q, are not composed of pairs of corresponding parall triangle ABC and Q; is the quadrilateral 4,B,By'Cy; sides AB, BC, and CA of @ are parallel to and have the same directions as sides AsBy, By Cy, and CyA; of Or, but Q has no side parallel to and hav ing the same direction as B,By' 107Side B,By' is part of a supporting line / for polygon Ox. Polygon Q has a supporting line / parallel to /, and having the same direc- tion. Line / passes through only one point of Q, that is, the vertex B. We then say that polygon Q has a degenerate side BB’ parallel to B,By' and having the same direction. Side BB’ is called de- generate because its end points coincide. The addition of this degenerate side reduces the present case to the preceding case, in which polygons Q and Q; consisted of pairs of parallel and similarly directed sides Then Q, = sO + 5:0; is a polygon A,B,B,'C, in which sides A,Bu, BBs, By'Cy, and C,Ay ate parallel to and have the same directions ‘as the corresponding sides of Q and Qj. The vertices As, Bs, By, Ca divide segments 44s, BBs, B’B:', CC, in the ratios,/s. We shall now apply this method to the general case. Suppose that we are given two polygons Q and 01. Suppose also that Q; does not have a side parallel to and with the same direction as side AiAux of Q. Draw the supporting line I’ for Qy parallel to Aidit and having the same direction. This line passes through only one vertex 4’ of polygon Q,, Now we say that polygon Q; has a de- generate side A/A’s,4 parallel to AsAs,1 and having the same direc- tion (both end points of 4,’'1,1 coincide with point 4"). We do the same for Q if it fails to have a side parallel to and with the same direction as some side of Os. By adding the necessary degenerate sides as shown above, we may say Any two polygons are composed of pairs of parallel corresponding sides with the same directions The analogous situation for polyhedra is used in Chapter 5. We shall now compute the area of a convex polygon Q with ver- tices As, 42, ..-, Ay (Fig. 114). We translate Q so that the origin O lies inside Q. The ith side 4,4,.1 has an equation of the form xeosa ty sina = hy where fi is the distance from the origin 0 to the side AvAiyx Ifthe length of AvAss1 is a, then the area of triangle OAvAL-1 is half the product of the length a; of the base A:Ai.1 and the altitude fy ata, = Sas Since the entire polygon Q may be divided into n such triangles, then its area, J(Q), is given by J(Q) = Area of OAA; 3 at Now consider two polygons Q and Qx (Fig. 115) with parallel and similarly directed corresponding sides. (As we have just seen, any two polygons may be so considered upon the addition of certain degenerate sides. This does not affect the computation of the area. because the length of a degenerate side is zero.) Denote the vertices of Q by As, do, ..., Aq, and the corresponding vertices of Qs by Av, Ag, + Aq’. Let a, and ai’ be the lengths of sides Avda: andAVA’ yx. (These sides are parallel and similarly directed.) Sides Aida and A/A’441 have equations of the form cos a + sin a = hy xeosa +y sina = hy. By formula (1) the areas J(Q) and J(Q;) are given by J(Q) = Sak, ° IQ) = @) ‘We shall now compute the area of the polygon Qe = 30 + Or. » polygon Q, has vertices Alo = sAy + 5A) and the length of side A\A, By the previous resul (9 = say + sya. The equation of this side is x cos a + y sin a = hy where Ii) = shy + hi. From formula (1) the area of Q, is Soa + 5a’) (shi + sib’) 3) HQ) =4 Sa liminaing the parentheses under the summation signin equa tion (9 and collecting the terms with, sand 2, we obtain @ By formulas (2) and (2’) the coefficients of s? and s,? in the right side of formula (4) are equal to J(Q) and J(Q,). We shall study in detail the coefficient of 5. We first prove the equality of the two sums appearing in the coefficient of si S aihe 1S ay = rT, hy = Drop perpendiculars from point 0 to the sides of polygons Q and 1. (The perpendiculars to a pair of parallel sides with the same direction will coincide.) For simplicity, we may assume that O lies inside both Q and Qs (Fig. 115), because it may always be made to do so by suitable translations of these polygons. Let Ci, Ca,.-- Gx denote the feet of the perpendiculars dropped from 0 to sides Ayda, AaAa, .. . Ans of polygon Q. Similar denote the feet of the perpendiculars to the corresponding sides of polygon Q;. Then we have OC, = hy, OCz = hay vee OC, = hy OCY = hy’, OCz = in’, OC = hi! By connecting each point Cj’ with the two neighboring vertices ‘Aj, Aces of polygon Q, we obtain the polygon A1Cy’AsC2'..AnCn'Ax bounded by the dashed line in Figure 115. Its area is “To prove this, observ that the area ofthe polygon is the sum ofthe sen erquadeloerals OlyCiide, O4sCrty, OAgCy ty The area of quadrilateral 04,Cy’As with perpendicular diagonals OCy’ nd. zds is hall the product of the Teng of these diagonal, or jah Sinilry, the aren ofthe next guadlatra is bay. and SO forth. Therefore the total area of our polygon is as stated. ‘We now construct the polygon 4y'C,dq'Co.. . Ay! Cad’, Whose boundary represented ia Figure 118 by thetine consisting of dos and dashes Ad above, we ean prove that the area of this polygon is Lewy ae‘The polygon bounded by the ordinary dashed line is obtained from polygon Q by adding to it the 2n triangles (shaded by hori- zontal lines) 1» CxCa'Aa, AaCy!Co, CoCa!Aay «+ AnCn Cay CuCy’ Ar The polygon bounded by the line of dots and dashes is obtained from polygon Q by adding to it the 2n triangles (shaded by vertical lines) AA: C1Aa'As, Aaa’ Ca, CoAs’As, «5 Ann! Coy CyAs Ars Notice that the areas of the triangles in the above sequences are pairwise equal. For example, triangles 4Ay/Cy and AxCy'C; have ‘common base 4,C,, and the vertices 4,’ and Cy’ opposite it lie on a line parallel to the base, giving the triangles equal altitudes. Therefore the areas of the two triangles are equal. Similarly the re~ ‘maining pairs of triangles are equal: CyAy'Ag and CyCy'Ap, AaAa' Co and A2Cy’Co, et. The areas of the figure bounded by the dashed line and the figure bounded by the line of dots and dashes are equal, because these figures consist of pairwise equal parts. It follows that - ani =4 $ al 7a 7 Dertsriox, The expression $ ah’ or LS ahi called he mixed area of polygons Q and Q,. It is denoted by J(Q, Os) ‘Thus, formula (4) takes the form J(Qs) = 8*I(Q) + 2ssxJ(O, Ox) + 1°J(Qs) 4) Notice that the mixed area of polygons @ and Qs may be obtained from formula (1) for the area of a polygon by using the lengths of the sides of one polygon and the lengths of the perpendiculars dropped from point O to the sides of the other polygon. If Q.= Qs, then the mixed area J(Q, Q1) of Q and Q; coincides with the ordinary area of Q. Incoursesin elementary geometry the area (or the circumference) of a circle is defined as the limit of the areas (or perimeters) of in- scribed and circumscribed polygons. The area of any plane convex figure © and the length of its boundary q are defined analogously (cf. sections 6 and 12), For each integer m > 3, we construct an in- 12 scribed polygon 9® with m sides (Fig. 116). As m, the number of sides, increases without bound and the lengths of the sides approach zero, the polygons Q approach the shape of figure © The areas J(Q) approach a limit which we denote by J(B) and define to be the area of figure O. We can also construct circumscribed poly- gons 0," about ©. As m increases without bound and the Q:( con- verge to ©, their areas J(Qy) approach the same limit J(8) Fig. 116 Let g and qx denote the boundaries of Q* and Q,". As n= co, the lengths 1(¢) and [(g.) of the perimeters 4 and qi approach a common limit /(q), which we define to be the length of g Thus J(Q) = lim J(Q™) = Jim J(Qs" 1g) = lim qu). 1@ = We shall not give the details of the rigorous basis necessary for these definitions. This would be done in essentially the same way as the area of a circle and its circumference are presented in elementary courses in geometry. ‘Now we shall consider the three plane convex figures $1 2 0) 2, O, B=D+ sn CHa =l ‘We shall construct a sequence of inscribed (or circumscribed) poly- gons Qt converging to 8, and another sequence Qs converging to 21. Then the polygons Q, = sO + 5,040 will converge 10 Ox Thus J(Q) = lim 1), T(Qy) = lim J(Qi), J(Ds) = lim J(Qs). 13On the other hand, from formula (4'), which is proved for all polygons, we obtain TQM) = $2F( OH) + 2853I(Q", O19) + s2F(Q4), (5) As n increases without bound, the coefficient of in formula (S) ap proaches J(B), the coefficient of s,? approaches (2), and the terin on the left appraches J(B,). The remaining term, containing 2s, ‘must also approach some limit, that is, the mixed area (Qt, Q,™) of Q and Oy approaches some limit as n —> co. It define the mixed area of convex figures Q.and By to be t write Taking the limits in formula (5), we obtain F(Ds) = 591) + 2s5,J(D, Os) + 5:91) 6) This looks exactly like formula (4’), which was proved for convex polygons only. Now we have shown that itis also valid for arbitrary plane convex figures. The concept of mixed areas turns out to be extremely fruitful We shall show in the next section that several geometric quantities associated with a convex figure © may be defined as the mixed area of © and some other figure. 28. APPLICATIONS ExampLe 1. Length of projection as a mixed area. Consider an arbi- tary polygon Q and a line segment / = 4B of length | (Fig. 117) Polygon @ has two sides with opposite directions parallel to /. (As above, these sides may have length zero.) Denote these sides by by and by, Segment / may be considered as a degenerate polygon L with two sides by and b/ of length 1, parallel to by and 6, and having the same directions. (One side is directed from A to B, and the other from B to A.) Any additional sides of this degenerate polygon have length zero, Polygon L has only two vertices, A and B. If'by’is a side of polygon L, then we denote its length by ay’, Now we have aa where k is different from i and j. Now consider J(Q, L). We know that HQL) =$ 5 ape, where py denotes the length ofthe perpendicular dropped from © to side by of polygon Q. Since all ay’ = 0 except a’ and a;', whict equal 1, we have 10.1) =i +) on Q parallel to Now draw supporting lines hy and fy for polygon Q pi segment. Sides 6 and b ie on these line, pis the distance from 0 to hy pj is the distance from 0 to hy, and py + p; is the distance Ddetween fy and fy. If m is a straight Bae perpend cular tof, then «+ pyis the length of the projection of figure @ on m. Fave proved the flowing propoion forthe ce that Qi iygon. In passing to limits, Q may be replaced by an arbitrary seme tigre, and We can prove the Vldtyof the propostion for convex figures. sth \ and mis a line Tarores 4. If lis a line segment with length | : perpendicular to |, then the mixed area of convex figure . and | is numerically equal to half the length of the projection of © on m.EXAMPLE 2. Perimeter as a mixed area. Now consider an arbitrary convex polygon @ and let Q; be a convex polygon circumscribed about a circle with radius 1 so that the sides of Q; are parallel to the corresponding sides of Q and have the same directions (Fig. 118). Fig, 118 Given polygon Q, itis always possible to construet a correspond- ing polygon Q; satisfying the above conditions. First, draw a circle center at O and radius 1. Then draw tangent lines to this circle parallel to the sides of Q and having the same direction. These tangents form the boundary of the desired polygon Q;. The mixed area J(Q, Q1) is given by LS ayy $3 oe. IQ, Ox) = where nis the number of sides of Q, ay is the length of the ith side, and pj’ is the length of the perpendicular dropped from O t corresponding side of Qs, Since the sides of Q, are tangent toa circl with radius | and center at O, then Therefore 6 ‘The right side of this equation is half the sum of the lengths of the sides of polygon Q, that is, half the perimeter of Q. Thus, we have Tuzonem 5. The mised area of convex polygons Q and Qs, where 4 is circumscribed about a circle with radius 1 and has sides parallel to the sides of Q and with che same directions, is numerically equal to half the perimeter of O. Using Theorem 5, we can prove the following: THEOREM 6, The mixed area of aplane convex figure O.anda R with radius | is numerically equal to half the length d of the bound- ary of Bs JB R) = 4a Proof, Here Dis an arbitrary plane convex figure, R a circle with radius I, and d the length of the boundary of £. Around © and R, construct sequences corresponding sides paral creases without bound, polygons 0 converge to O, and Re to R. By Theorem 5, HQ, RO) = where d® is the length of the boundary (perimeter) of creases without bound, d® approaches d and J(0* proaches J(©, R), and we have the desired result. 29. SCHWARZ INEQUALITY; OTHER INEQUALITIES In the next section, we shall prove an inequality relating the areas JQ), J(Dx), and J(B, O4). First, we must prove certain other in- equalities, which, in turn, are connected with some inequalities concerning the coefficients of a quadratic equation. Tf we are given the quadratic equation att + 2bt + ¢ = 0, a then its oss hase he form L(cbe yam v(i) If 8 — ae <0, the roots of equation (1) are complex. There is no real root; that is, the expression ar + 2bt +e oy does not become zero for any real number f (8) 1f 8 — ac = 0, equation (1) has equal real roots = —b/a. Expression (2) equals zero only if = —b/a. Using these facts, we immediately prove the following, SckWARz INEQUALITY. For any pair, y and y, of functions of x we have the inequality [ Ronee]? < [ree [P nes, °% in which equality holds if and only ifthe functions are proportional that is, via) = ky(x) _for some constant k. Proof. For any real number ¢ we have Slo + yntden [Fyre 426 fiprdes 2 fyi The expression in the integral on the leftis the square of some fune- tion. Therefore the integral is nonnegative, This integral is zero if and only if YQ) + Mn@) =0 for every x between x’ and x”. In this case C= v1) and the functions y and yy are proportional, with —1 as the coef= ficient of proportionality. If the two functions are-not proportional, then the integral on the left is positive for every z (assuming x’ < x”). Letting 4 [Ordna [marae [Stdeae, we see that if functions y and y; are not proportional, then the expression, a4 2b + ef is not zero for any real number t. Therefore, we have be ae <0. But if the functions are proportional, then this expression becomes zero if and only if —r is the constant ratio of these functions: = 20), “= ye) In this case bt — ae = 0. ‘Thus we always have the inequality b? — ac <0, thatis, be < which is the inequality [fon ax] < fide [Pts in which equality holds if and only if the functions y and ys are proportional. The proof is complete. Now let us prove some additional algebraic inequalities. Itis always true that a2 — Lab + b = (a — by? > 0, from which we obtain a? +B? > Dab, @ ‘We have proved Taorem 7. The sum of the squares of two real numbers, a and 5, is not less than owice their product. Equality holds when a = b.‘TuwoneM 8. For four real numbers, a, a1, b, and by, we have aa, — bby > VB Va BF oO ifall differences are nonnegative. Equality holds when aby — a,b = 0. Proof. (aay — bb;)® = aay? — aaybb, + b*b;*, — byt) = aay? — aby? — 48d? + DA. ares of the numbers ab; and where equality holds if ab, — ayb = 0. From this it follows that (aa, — bby)? > (a? — b8) (ax? — bi), and our theorem is proved. ‘Tonea 9. For any pair of functions y and ys, with x as the in- dependent variable, and any pair of real numbers, a and a3, we have the inequi aay — fF yr de Equality holds if and only if the func- 10 the numbers a and ay; differences are nonneg: tions y and y1 are proport YG) _ ai ye) in (6) we must have necessary that both and (9), and for this HO g vey where k is some constant, and by @), ®)30. RELATION BETWEEN AREAS OF Q, Qi, AND Qs; ‘THE BRUNN-MINKOWSKI INEQUALITY BRUNN-MINKOWSKI THEOREM. Suppose that we are given plane convex figures BO, Basta Gus hss and denote their areas by I(),J(B1), and JB). Then we have the inequa VIB) = s VIB) + 81 VER) ) holds if and only if © and Dy are homoth larly arranged. Inequality (1) was first proved by G. Bruna, G. Minkowski de- termine’ when jt becomes an equality. We remark that inequality (1) is valid not only for convex figures but also for arbitrary figures in which equali that rand si Proof, We shall prove first that if © and ©) are homotheti, the equality holds in (1). If © and O, are homothetic, then 0. is homo- thetic to both of them, Thus if 8; is congruent to KO, then Dis sD + kD = (6 + 51k). se conditions Ha) = KID), Bs) = 6 + Hk°IO) VICE) = (5 + sik) VIB) = s VIED + kV) = 5 VIB) + VI ‘Thus, we see that formula (1) is satisfied for homothetic figures, and equal We shall show that inequality equality (1) is equivalent to the following JB, &) > VIEITE For the proof, recall formula HDs) = 89D) + 2951F(Q, 1) + 51°7(Bs) ® By squaring both sides of inequality (1), we obtain IQs) > 2D) + 2861. VTBY TER) + 72). O @ ) of section 27: By means of formula (3), inequality (4) may be obtained from (2) and conversely. We shall undertake to prove (2) and hence (1 ‘There are many proofs of the Bruna-Minkowski theorem. We shall use the geometric method of Frobenius (1849-1917), a German mathematician, Consider two arbitrary convex figures © and O; (Fig. Around these figures construct circumscribed triangles ABC and ‘:BxC; with corresponding sides parallel If Qs Ptah 6+H=b5n 20, then 1) = SAABC + ,4ABiCx Here, AA,BC, smscribed about O, with si parallel to the sides of the other two triangles (and therefore to both of them). ‘Through a point D on the boundary of figure © draw a support ing line. It intersects triangle ABC at points M and N. Segment DM, having the direction of positive motion around 9, is called a positive 3rive supporting segment. Let ‘DiM, and D,M, be the positive supporting segments for figures Or supporting segment, and DN, a nega’ and Q, parallel to DM, These segments make some angle « with the suaxis, We denote the length of DM by 1(a), the length of DMs by h(a), and the length of DyMz by la(a). We have 1,(a) = sl(a) + sih(a). o Consider the set of all positive supporting segments for figure © (also for © and 1,). It fills the parts of triangle ABC that lie out- side ©. We denote it by (AABC — ©). (Analogously, the positive Supporting segments for figures Q; and ©, fill the regions (GAxBiC; — B1) and (A,B,C, — 84). We wish to find the area of region (AABC — 9). To do so, at point D on the boundary of © construct positive supporting segment OOM. Draw a line DM’ to some point M’ very close to M on triangle “ABC. We have constructed an infinitesimal triangle MDM, with an Jnfinitesimal angle da at vertex D. As before, 1(«) denotes the length T DM, The area of this infinitesimal triangle is given by #/%(a) da. The area of region (AABC — ©) may be considered as the sum of the areas of stch infinitesimal triangles. We represent the area by the integral 1 [rae Let us denote the area of triangle ABC by J, of triangle AsBiCt by Ji, and of triangle A,B,C, by Jy. Then we have JB) as-4 [7 P@ de © Analogously, (Ba) = ts — EJ HD de a and SOs) = Fe =F [2 12C) de ® Because triangles ABC, 41B:C}, and A,B,C, are homothetic, we have Ie= VT + ivi) o From formulas (9), (5), and (8), it follows that qe 4 [ vez [sl(a) + sils(a)}? dav H(Qs) = (syT +5 = 9 + QV + 3 =F L202) + 2ssstop(0) + 52820] de =" (ayh(a) dal {1-4 fm] + 20f safe (On the other hand (from formula (5°) of section 27) I(Ds) = s24(B) + 258J(B, Oh) + 7°V(Ds), in which the coefficient of s? is the area of O, the coefficient of s1* i the area of Q;, and the coefficient of 2ss, is the mixed area of O and Oy. Therefore, JB, D1) = VII — $ JF) 400) ae. (ao) 140) dal (6) of section 29. Let the numbers a, a Now turn to inequali be defined so that Je@ and Jy =a, hence, VIR = aay ‘Then replace yx, uC) and x in (6 of seation 29 by GD, MQ) and a. Thus, . “ VIF — $ JP" He) ha) de 2 /fr-F re) dads: — $2" 4209 ae .d from (6), (7), and (10), it follows that J(O,.2:) > VIET. «2 ‘Thus, the Brunn-Minkowski inequality is proved.It remains to be proved that if the equality in (12) holds, 2 and Q, are homothetic, For equality to hold in (12) it is necessary that for all a Wo) _ VT h@~ vi ‘We shall show that this happens only when © and © are homo- thetic, Note that the ratio ofsimilitude of the similar triangles ABC and A:B,C, is VJ/ yJ«. Denote this ratio by k. Then the ratio of the lengths of corresponding sides of these circumscribed triangles is k; the ratio of the lengths of parallel positive supporting segments for figures 8. and © is also k. In all our discussions, if we replace positive supporting segments by negative supporting segments, then Wve obtain the sane result: the ratio of the lengths of parallel nega- tive supporting segments is also equal to k. ‘Now let MN and M,N; be segments of parallel supporting lines for © and O, cut off by the boundaries of triangles ABC and AxBiCs We have MN = MD + DN, MNy = MiDs + DiNi where MD and M;D; are positive, and DN and DyNy are negative supporting segments. Since MD = kM,D:, DN = kD.Ns, we have MN = kMNs Now consider the figure Op = KOs; around it is the circum seribed triangle A.AgBaC2 = KAAxB,C, Segment MoNz = kMyNy isasupporting segment for O. with end points lying on the boundary of triangle 4,BzCa. The corresponding sides of triangles ABC and “aB2C2 are equal and parallel. Thus, the triangles are congruent. The triangles cut off of ABC and AaB-C, by segments MN and MoN2 (for example, triangles AMN and 4pMzN>) are also congruent. They are similar because their corresponding sides are parallel, and con- gruent because sides MN and MpNo = kM,N; are equal. Triangle {5B2Ch is obtained from triangle ABC by a translation. The same translation applied to triangle AMN yields triangle AzM2No, and applied to supporting line MN, it yields MzNs. Since the paiz of sup- porting lines MN and M,N is arbitrary, we conclude that any Supporting line for figure Dz is oblained from a supporting line for gure © by applying the same translation. Since figures © and Qe = kD, are determined by their support- ing lines, © and Oz must be related by the same translation as their supporting lines. Thus, © is obtained from ©; by multiplication by i followed by a translation; that is, figures © and ©, are homothetic. ‘This complétes the proof that equality holds in the Brunn- Minkowski inequality only for homothetic figures. Remark. The Brunn-Minkowski theorem is easily generalized to redimensional space. For example, for three-dimensional space, we feplace the square roots in inequality (1) by cube roots and use volume instead of area. 31, RELATION BETWEEN AREAS OF PLANE SECTIONS OF CONVEX SOLIDS Suppose that we are given two plane convex figures © and ©: in two parallel planes P and P; (Fig. 120). We can form a solid 7 by connecting every point q of igure © with every point qs of figure 5; with straight line segments. The set of all these segments forms some solid 7. (In Figure 120, figures © and O are represented by parallelograms that are not similar) Consider the section D. of solid T made by plane Py=sP +5:P1(5-481 =1i 5,51 > 0). Plane Py is par- alle to the planes of figures ‘Band ®, and divides the distance between them in the ratio s1/s. Each seg- ment gg: of solid 7 (q be- Tongs to ©, qi to 93) in- tersects plane P, at the point which divides the segment in the ratio s/s, that is, at the point qe 99 + 51h Denoting the set of all points g, by Os, we have Oy = sO + sO. 27In the preceding section we proved the Bruan-Minkowski theorem for figures ©, Oy, and B. lying in the same plane. But every step of the proof is equally correct for ©, Dy, and ©, lying in dif- ferent but parallel planes. Thus we still have the inequal VIE) = 8 VIB) + 5vTEH) o Also, inequality (1) becomes an equality if and only if figures © and ‘Dy are homothetic. If O and D; ate homothetic, then the solid Tis easily seen to be either a cylinder (Fig. 121) or a frustrum of a cone (Fig. 122) (or simply a cone, if one of O and © isa point (Fig, 123) For simplicity, we use circular cylinders and cones in our ilustra- tions, but © and ©, may have other shapes. Fig. 121 Fig, 122 Fig 123 Now suppose that we are given a convex solid Kand some straight Jine which we shall consider as the x-axis. We shall study the sec- tions of K made by planes perpendicular to this axis (Fig. 124). Let us take two points x, x1 on the x-axis and an arbitrary point x, between them, such that mar tom(s $5 = 55120. Denote by 0, Q:, and Q, the sections of solid K made by the planes P, ,, and P, perpendicular to the x-axis at points x, xs, and x,, Note that this definition of Q, is different from the one given before. TueorEM. The areas of sections 0, Qx, and Q, as defined above satisfy the inequality VIO) = sv) + av), 2 inwhich equality holds ifand only ifthe pat of solid K between planes P dand P is a cylinder or a frustrum of a cone. 18 Fig, 124 Proof, Let T denote the figure composed of all line segments necting points of section Q with section Qs Since Q and Q, are con- tained in convex solid K, all segments connecting their points belong to K. The solid T, composed of these segments, is contained in K. Let O, denote the section of T made by plane P,. Since Tis contained in K, then @, is contained in Q,. Hence, the area of Q, is greater than or equal to the area of F(Qe) = Ds) ‘The Brunn-Minkowski inequality is already proved for solid 7 VIG = s VID) + 1 VIO) eo) Therefore, VIG) > svVI@Q) + nVO)- Equality holds in (2) if and only if both the following c« are satisfied: (1) The part of solid K contained between planes P and P, coincides with solid T. (2) Sections Q and Q; are homothetic. ‘These conditions are satisfied whenever the part of K contained be- tween P and P; is a cyclinder or a frustrum of a cone, and the theorem is proved. 19We also have the following: Corottary. If the sections of convex solid K made by parallel planes P, Py, and P, have equal areas, then the part of solid K be- ‘tween planes P and Py is a cylinder. Proof. The preceding theorem tells us that this part of Kis either a cylinder or a frustrum of a cone. But since the areas of the sec- is are equal, it must be a cylinder. 32, GREATEST AREA THEOREMS The Brunn-Minkowski theorem has some interesting con- sequences. L’Hunuire’s mcorest. Of all convex polygons with sides having given directions and with perimeter of given length |, the one with the greatest area is circumscribed about a circle Proof. Let Q be a polygon with perimeter of length /, and let Qy be a polygon with sides parallel to the sides of Q, and circum- scribed about a circle with radius 1. Let S denote the area of Q and S; the area of Q;. By the Brunn-Minkowski inequality (formula (2) of section 30), we have IQ, Q1) > VOY TOD = VSS oO where equality holds if Q and Q, are homothetic. The mixed area J(Q, Q,) of polygons Q and Q; is numerically equal to half the length of the perimeter of Q (section 28): (Q, Os) = $1 From this and inequality (I) follow Lie VO, le>s.s, qi 2 SS: ck and WO) =S< 4 ® 130 in which if Q is not circumscribed about a circle (that is, not homo- thetic to Q,), we have the strict inequality a IQ <3, But if Qs circumscribed about a circle, inequality (2) becomes the equation ee MQ) = ¥5- ‘Thus, the maximum area of polygon Q is 22 and is atined i O's circumsoribed about a cre (hat iss omothetie 10 02). TaeoreM 11 (IsoPERIMETRIC PROBLEM). Of all convex figures with boundary of given length I, a circle has the greatest area. Proof, Let Di be a convex figure with boundary of length ! and let Q; be a circle with radius 1, Denote the area of Eby S. The area of Oy is = and its boundary has length 2s. ‘By the Brunn-Minkowski inequality we have ID, O1) = VIB) TD = v7. which becomes a strict inequality if 8 is not a circle. ‘The mixed area J(O, Qs) is numerically equal to half the length / of the boundary of © (Section 28); so we have lis vy ale ve, from which we obtain <2, ssf @ Equality holds in (3) if and only if © is a circle. Thus, of all convex figures with boundary of length J, a circle has the greatest area We shall prove later (section 40) that inequality (3) holds for all, (not only convex) figures and that the circle has the greatest area among ail figures of given perimeter, 1