AN ABSTRACT OF THE THESIS OF Murari Prasad Ojha May 27, presented on Mathematics in Doctor of Philosophy for the degree of 1982 Smooth Approximations anduermassintegrals Title: Signature redacted for privacy. Abstract Approved Harry E. Goheen Let [a,b] be a closed bounded interval of the real line, Eq the q dimensional Euclidean space, and y curve. If to c [a,b], y(to)), and to = : [a,b] let to = sup {t E[a,b] C [a,b] inf ft : t to : a continuous E t to , y ([to,t]) = , y([t,t0]) = y(t0)}. Let P be a polygon inscribed in y corresponding to a partition Q : a = tO < tl . < tn = b of [a,b] . Let 9(ti_iti,titi+1) = the smallest non-negative angle between the directions - y(ti_1) and -y(t1) - y(ti), with the proviso that 0(t1_1t1,titi+1) = 0 whenever y(ti_1) = y(ti) or y(ti) = y(ti+1). Let K(P) = max {G(t1_1ti,titi+1), i = 1,2,...,n-1} p(Q) = max {t; - t1 , 1 = 1,2,...n} . , and A curve y can be smoothly approximated by inscribed polygons if for every c such that in the open interval (0, p(Q) < d K(P) < implies 2 there exists a d > 0 ) c . The following approximation theorem is the main result in this dissertation. Approximation Theorem. A curve is regular if and only if it can be smoothly approximated by inscribed polygons. Besides this approximation theorem, the dissertation also contains research work on integral geometry. The measure of r-planes that intersect a convex set C in Eq can be exis a constant depending on r, pressed in the form x W (C) where r r Ar and Wr(C) is the rth quermassintegral of C. It is shown that it is possible to extend this result to a non-convex set T of the type T = U C. , where the C. are convex sets such that C. n 1 > 1, provided W (C.) and W (C r 1 n C14.1) are known. cj =0 for By taking r C. as convex tubes, the calculations for the quermassintegrals are 1 carried out for the case q = 3. SMOOTH APPROXIMATIONS AND QUERMASSINTEGRALS by Murari Prasad Ojha A THESIS submitted to Oregon State University in partial fulfillment of the requirements for the degree of Doctor of Philosophy Completed May 27, 1982 Commencement June 1983 APPROVED: Signature redacted for privacy. Professor ' Mathematics in charge of major Signature redacted for privacy. Head of Department of Mathematics A Signature redacted for privacy. ,PI ] () Date thesis is presented 11711 Typed by Bonnie Shula for <g (41 ) Murari Prasad Ojha ACKNOWLEDGEMENTS I wish to thank Dr. Harry E. Goheen and Dr. J. Wolfgang Smith for moral encouragement and valuable suggestions. I am especially grateful to Dr. Smith for his help with the proof in Chapter II that smooth approximation implies regularity. I also wish to thank Dr. Michael Papadopoulos for constant encouragement, reassurance and moral support. Finally, I owe my eternal gratitude to my family for the love and patience they expressed while I was working on this research. Table of Contents Introduction 1 Smooth Approximation of Space Curves by Inscribed Polygons 4 Measure of Lines that Intersect a Non-Convex Set--17 Quermassintegrals of the Intersection of Two Convex Tubes 32 Bibliography 41 LIST OF FIGURES Page Figure A convex tube T(r,o). 17 A tube T around two intersecting line segments. 18 Number of (line) segments, x(AnT), in the intersection of a line A with tube T (Figure 2). 19 A polygonal tube. 21 Intersection of convex tubes around two intersecting line segments OA, OB. 26 of tube around OA, the plane V of inter- Cylindrical surface C1 section of the cylindrical regions of convex tubes T1, T2 respectively around line segments OA, OB, and the plane U through 0 with OA perpendicular to U. 27 Plane U of Figure 6. 28 A convex set D of intersection of two convex tubes. Parallel convex set D + pB, where p > 33 0 and B is the unit ball. 35 The region (D + pBAD below the plane V. 36 x)-plane where Intersection of the region of Figure 10 with (x3' A is a line through the origin 0 on the plane U. 36 The planes of intersection (as in Figure 10) as A rotates 37 on the plane U. Figure Page A plane as shown in Figure 12 (or, as in Figure 11). 37 Part of intersection of (D + pB)\D and the plane U. 37 The angle between a line x on U, and the line of intersection of the plane V and the plane (x3, x). 39 SMOOTH APPROXIMATIONS AND QUERMASSINTEGRALS CHAPTER I Introduction The fact that a continuous curve y : [a,b] Eq can be uniformly approximated by inscribed polygons has played a basic role in mathematics. For example, it underlines the notion of rectifiability and the definition of length. Moreover, in the context of numerical analysis a curve is generally approximated by a finite sequence of Now the following question is points, which is in effect a polygon. a natural one: if y is not only continuous, but also smooth in some sense, and if one considers a sequence of inscribed polygons with mesh tending to zero, how does the smoothness of y reflect itself in the behavior of the polygonal sequence? Is there some property of the sequence iteself - which a computer can see - which is characteristic of the fact that the limit curve is smooth? The main result of this dissertation constitutes an answer to this question. smoothness to mean regularity, the desired property mating polygons turns out to be the following: given If one takes of the approxiE > 0 there exists a 6 > 0 such that, for every partition of [a,b] with mesh less than 6, the angle between successive edges of the corresponding inscribed polygon shall be less than E . Only one should add that in as much as y need not be locally oneTto-one, the mesh of partitions 2 This, then, is the new concept must be defined in an appropriate way. to which one is led, the notion of "smooth approximation by inscribed polygons." Precise definitions and a proof of the Approximation Theorem will be given in Chapter II. Though this is undoubtedly our main result, it is not one which we had envisaged from the start. As it turns out, we began our research by investigating two problems pertaining respectively to differential and integral geometry of convex sets. The first dealt with the approximation of q-convex curves by inscribed polygons. In particular, we were interested to find out whether q-convexity was sufficient to guarantee certain good properties of the approximating polygons. Eventually, however, it became clear that q-convex- ity had little to do with the questions at hand, and this recognition prompted us to consider the problem of which we have spoken before. As concerns the integral geometry of convex sets, on the other hand, our work has led to a number of results, which will be set forth It is well-known that the measure of r-planes in chapters III and IV. intersecting a convex set C in Eq can be expressed in the form where xis a constant depending on r and Wr(C) The question arises: quermassintegral of C. extended to the case of non-convex sets? x r Wr (C) is the so-called rth can this formula be We show that this can be done for sets of the form T = U Ci, where the Ci are convex sets and C. n c.j = 0 for li-j1 > 1. This in turn leads to the problem of evaluating quermassintegrals for the intersection of convex sets, 3 which in general is a rather difficult matter. We have succeeded in carrying out the calculations for the cases r = 1, 2 and q = 3 by taking C. as convex tubes. 1 CHAPTER II SMOOTH APPROXIMATION OF SPACE CURVES BY INSCRIBED POLYGONS Definitions Let [a,b] be a closed bounded interval of the real line, E the q dimensional Euclidean space, and Sq-1 the q-1 dimensional unit sphere. A curve is a continuous map y A polygon P (in E [a,b] : is a sequence ) a0' a1 ' ...., an and the line segment joining the points ai_1 inscribed in P is < tn= b such that y . of The points are called vertices pointsinEqsuchthataa.1' 1+ of P. E ai is called an edge , if there exists a partition a = to< t1 = ai For every to I (t X' ) 0 y(C). = 0 [a,b], to #t t , If xo x'0 , let u[t 0 Let t [a,b], E , t'] denote the unit 0 0 vector in the direction from xo to xO let xo = , . and define 0 = sup { t6 = inf { t E[a,b] E[a,b] t Then a < t- < t O. (i) < t 0 t]) = 1(t0) = xo 1 : t to , Y([to : t to , Y([t, to]) , = 1(t0) = x0} b. 0 Suppose t b. A unit vector ur is a right tangent if,for every sequence tn E[a,b] such that tn > to lim n lim = t+ t n 0 ' . n u[t' 0 t] n = u . r , y(tn) to y at xo , to and < 5 If t+ , 0 then y has no right tangent at to. Suppose t- > a. (b) A unit vector u is a left tangent to y at t 0 0 1 if, for every sequence tn lim tn = t0 lim u[t,tn] = - u1 ' n c[a,b] such that tn < t y (tn) xo, and . n If t- = a, then y has no left tangent at to. 0 Note: If (iii) Suppose a< ur or uI exists, it is unique. t0 A unit vector u is said to < b. t0 < 0 be a tangent to y at t if and only if 0 the right tangent ur exists at to the left tangent ul exists at to u = ur = ul , , and . Suppose y has a unit tangent u(t) at every t y : [a,b] by setting S admits continuous tangents if (D) Let 0(t1t2 between u[ti O(tit2 O(t/ , , , We = u(t). We say q-1 % define a map y E[a,b]. is continuous. t3t4) = the smallest non-negative angle , t2] and u [t3 t3t4) = 0 if t2t3) = g(t2t3 y , 1(t1) = , t4] y(t2) or 1(t3) = y(t4), t1) = the smallest non-negative angle between u(ti) and u[t2, t3], 0(t1 ,t2t3)= 0(t2t3 9(t1 , , t1) = 0 if y(t2) = y(t3), t2) = the smallest non-negative angle between u(ti) and u(t2). 6 If Q: a = t < t < 0 = b is a partition of the interval < t 1 [a,b], and P the corresponding polygon inscribed in y, let K(P) = max and = max p(Q) 0(ti_1ti, titi+1), i = 1,2,...,n-1 } { {t'i - , i = 1,2,...,n . A curve y can be smoothly approximated by inscribed polygons if for every E in the open interval (0, -TT ) there exists a 6 > 0 such that p(Q) < 6 implies K(P)< E. A rectifiable curve is regular if it is C1 with respect to (E) arc length parameterization. Note: Usually a regular curve is defined (for instance, see [1], page 6) as follows: y'(t) 0 for all t A Cl curve y E [a,b]. [a,b] : Eq is regular if Our definition concurs with this definition because any C1 curve y : Eq with y'(t) [a,b] rectifiable, and C1 with respect to arc length. 0 is However, unlike the usual definition, our definition is independent of parameterization. We give an example of a curve which is regular, but not regular with respect to a given parameterization. Example Define y: [-2, 1] E2 by y(t) = (- t2, t3), -2 = (t2, t3), 0 0, t t 1. Here y'(0) = 0, and so y is not regular in the usual sense. s(t) = ftlly1(T)11 dT. f -2 /4T2 + 9i4 dT. -2 Solving this equation for t in terms of s, and setting y(t(s)) = ; (s), we find that Let 7 ((4Q)3/2 ((40) 3/2 - 27s)213 -4 27s)213 -4 - 3/2 - -;,(s)= ( 9 9 (40)3/2-8 0s 27 (27s - (40) 3/2 + 16) 2/3 - 4 ( ( (27s - (40)3/2 + 16)2/3 -4 9 9 (40)3/2 -8 (13)3/2 + (40)3/2 - 16 s 27 . y is C 1 3/2 27 with respect to s, and so y is regular. Approximation Theorem. A curve is regular if and only if it can be smoothly approximated by inscribed polygons. I. First we show that if a curve is regular then it can be smoothly approximated by inscribed polygons. Let yr : [a,b] Eq be a regular curve, and let y denote the arc length parameterization of yr. : [0,L] It suffices to show that y can be smoothly approximated by inscribed polygons. For suppose that for every E in the open interval (0, there exists a 6 > 0 such that if 2 p(Q) < 6 then K(P) < E , where Q is a partition of [0,L], and P the corresponding polygon inscribed in y continuous function f: [a,b] [0,L] . There exists a monotone such that yr = y o f. Moreover, by uniform continuity of f, there exists a 5 0 such that, for every partition Qr of [a,b] with p(Qr) '< 6r, one has p(Q) < Let Pr be the polygon inscribed in yr corresponding to Q. 6. 8 So if p(Qr) < Sr then Let y length s. : Eq be a regular curve parameterized by arc [0,L] E [0,L] For every s, s* y(S) - y(S*) - h y'(5*) , s* s let o(h) = , , where h = s - s*. Given Lemma 2.1 K(p) < 6 (Pr) = > 0 there exists a 6 > 0 E 110(h)11 then such that if 0 <lhl < 6 < 6 . Ihl Proof of lemma 2.1 (yi(S), y2(5), , Let (5) = yq(s)). Then Y(S) 11o(h)11 0 and y > e Yl(s*)11 Y(54`) = 1h1 / [ yl Yl(s*) (s) - Since given E 72 (s*) yi y E Cl [0,L] If we choose 6 = 6i min(61 + C1 > 0 there exists 6.> 0 such that if 0 (S) -Yq (S*) * Y.(s) 1 y(s) - So yi(s*) then , y!(s*) 1 h 11(S*)] [0,L] for each i = 1,2,...,q. 62 then for each i = 1,2,...,q, if 0 <1h1 YiSs) - 2 < C FIT < 6 then So given 110(h)11 E > 0 there exists 6 > 0 such that if 0 <1h1 < 6 then < Ihl 62 E 9 Now to show that y can be smoothly approximated by inscribed polygons we need to show that given 6 > 0 such that for every so, sl, s2 < 12. 1s/ - s21 = g(si , < in (0, there exists 6 [0,L] with so <,s1 < s2 E 6 then 9(s0s1, s1s2) < E. Let , if 0 a = O(sis2, s1), and soy. By the triangle inequality on Sq-1 , we have a+ g(sosi, s1s2) (2.1) We have cos a = u(s1).u[sl, s2] Y(Si) Y(S2) = (Si).yI Y(S1)11 IlY(S2) hy1(s1) + o(h) 2 + 2hyl(s ) .o(h) + llo(h)112 1 where h = s2 - s 1 h + yl(s1).o(h) h\,/ 1 + 2 yi(s1). (1J0(h)11 0(h) h ) o(h) 1 + (s 1 ) (2.2) + 2 y'(s ). 1 o(h) 110(h)II ( )2 10 Let h = . By lemma 2.1, given 110(h)11 h El > 0 so that we have in (2.2) E1 2 1 < h > 0, we can choose E > 0 such < 61. ly.(so. o(h) Given 6 1 that if 0 < h < 6 then 11)(h)11 Also > 0 there exists E s2 - s1 - 1. cos a < 2 Given E in (0__) 0 so that a < , we can choose 62 > 2 That is, given there exists 6 > 0 such that in (0, E 2 if 0 < Is, - s21 a< < 6 then (2.3) Similarly if 0 < (so - then 13 < (2.4) 2 Substituting (2.3) and (2.4) in (2.1), it follows that given in (0, ) there exists 6 > 0 such that if 0 0 < 1s1 - < 2 s11 < E 6 Is0 - s21 < 6 then O(sosi , s1s2) < E. Hence a regular curve can be smoothly approximated by inscribed polygons. II. Now assume that y can be smoothly approximated by inscribed polygons. Let to We will show that y is regular. E [a,b]. We first show that y has a tangent at suffices to consider the case where y([a,t0]) y(b). It to. y(a) and y([to, b]) 11 Consider any sequence y(t ) and lim sequence u[t = t t n n ÷ 0 E[a,b] such that t, > t+ tn y(t , 0 H ) 11 By compactness of 5q-1, the . 0 t ] has a limit point , ur. Similarly if y(tI) tr'il and y(to) m is any sequence in [a,b1 such that lim = t- then the sequence u[t t' m 0 < , 0 t-6 ti] has a m limit point - ul. In order to show that y has a tangent at t , it suffices to show 0 that ur = ul. Suppose ur u/ and u C Let ip be the smallest non-negative angle between Let 0 = 0(t'tt t ). Then 4, # 0. . in (0, ul. ) ff2 m 0' so that E < . 0 n We may choose By the definition of smooth approximation, there exists a 6 > 0 such that if 0 <It -tI< 6, n 0 < itm - tol < 6 then 0 < E . 0 But lim 0 = 4), which is a contradiction. We now show that y admits continuous tangents. It suffices to 7 2 ) there exists 6 > 0 such that if 0 show that given E in (0, < It - t01 < Lemma 2.2 t, to < E. 6 then 9(t,t0) Given E in (0, ) there is a 6 > 0 so that for every < 6 then 0(t,t0t) < E [a,b], if 0 < t - t and 0(t o, tot) < E. 0 Proof of Lemma 2.2 Let 2 > 0 and determine the corresponding 6 > 0 from the definition of smooth approximation. Let 0 < t - to < 6. 12 If y (t) = y (t y(t). ) then g(t,tot) = g(to, tot) = O. So assume y(to) Then t- = inffTE [a,b] t, y([T, t]) = y T : (0} > to and there is a sequence tn such that to < t< t-, y(t), and y(tn) t t-) < , g(t,ut n nllm. 27-. By our choice of 6, tn = t-. Now let n Then g(t t, co. n c 2 Likewise, if < E. t+ = SupfTE[a,b] 0 : T , t0 lim 0 g(t+tm' tm = 1(t0)} < t, tm < t, , t 0 co liM M As before g(t tm = tl". T]) Y([t0 there is a sequence tm such that t m = 0(tot-, t-) 0 = 0 g(t+0 t+ 0 ' t) = < E. 2 co Lemma 2.3 77 Given E in (0, ) there exists 6 > 0 such that for 2 every t, to E[a,b], if It - tol < 6 then g(t, to) < Proof of lemma lemma y (to), and y (tm) 2.2. Let It - 2.3 toI Let E 2 E. as in > 0 and choose 6 > 0 < 6. If t > to g(t,to) If t < t0 g(t,to + g(t t,t ) 0 0 < 2 2 . 13 g(t,tto) + g(tt0,t0) < g(t,t0) 2 We now show that y is rectifiable. Given Lemma 2.4 E in (0, ) there exists 6 > 0 such that for [a,b], if 0 < It - every t, tl, t2, t < 6 and t < t1 < t2 < t' c. then 0(tti,tit2) < Proof of lemma 2.4 Given > 0 choose E 6 > 0, the smaller of 3 the delta in the definition of smooth approximation and in lemma 2.3. Then g(tv,t) + o(t,ttl) + e(tt1,t1t2) < e(tti,t1t2) 1r-+ E 3 Given Lemma 2.5 every to, tn E [a,b] ) there exists 6 > 0 such that for in (0, E if A Sn < sec 0 < tn - to < 6 then E , where Axn = y(tn) - 11Axnll y(to), and ASn = length of the inscribed polygon with vertices at y(t ), y(t1),..., y(tn), where to < t1 < < tn. 0 Let H. be the hyperplane through y(t) Proof of lemma 2.5 such that 13i is normal to Hi. AXn Let ai = IlY(ti) - Y(ti_011 = distance between the hyperplanes H1..1 and Hi 9. =g(t t 1 0 n' t. 1-1 t Ito - and ) i By lemma 2.4, given if 0 < , E in (0, tn1 < 6 then Oi < E, ) i = there exists 6 > 0 such that Hence = 14 aj = sec gi < sec Axn Let u = (2.5) . Then . IlAxn11 l[y(t.) - y(t. = 0 1- 1 <E< O. )].ul = [y(t) - y(t. 1- )].0 because Consequently, . 2 Ax = n [1(t.) - (t. E )] , and 1-1 i=1 E[1()- y(t t.1 )] . i-1 i=1 u = E 1lAxnII = Ax u = 13. i=1 ? Thus, A Sn FlAx < sec E 2 1 + + fi 2 1 , by (2.5). + fin y is rectifiable. Lemma 2.6 Proof of lemma 2.6 corresponding polygon. Let Q be any partition of [a,b] and P its There is a fixed partition Ql such that successive partition points of Q/ differ by less than the d corresponding to E = 1 in lemma 2.5. P* denote the corresponding polygon. L(P) L(P*) Let Q* = Q U Ql , and let Then L(Pi) sec 1 by lemma 2.5. where L(P) is the length of P. Thus, y is rectifiable. Finally we show that y is Cl with respect to arc length. Let a < t < b. 0 E[a,b] so that Consider any sequence tn tn > t0' 15 y(t ) y(t ), and 0 lim Let Axn = y(t tn = to. ) - y(to), and n ASn = arc length measured from 1(t0) to y(tn). Since y admits continuous tangents by lemma 2.3, it suffices to show that Axn lim where - u(to) u(to), n is the unit tangent at , so to ASn that y is C1 with respect to arc length. By lemma 2.5, given E in (0, ) 2 0 < Ito - tni there exists 6 > 0 such that if < 6 then A IlAxn1I 1 AS 11 > ASn where ASn cos E. ASn is the length of any inscribed polygon beginning at y(t ) and ending at y(tn). Since y is rectifiable by lemma 2.6, there exists a partition Q of [to, tn] so that if ASn is the length of the corresponding inscribed polygon, then AS n > 1 - . ASn Hence if 0 <Ito - ti < 6 then IlAxnll 1 >(1 >. - E) cos ASn lim n 11AXnII co AS = 1. (2.6) 16 We have 111(tn) - y(to) Axn ASn u [to, tn] ASn IIIAXn11 _ u[t0 ,t ] ASn Axn lim lim = -9- fl u [to, tn] = u(to), by (2.6). n ASn A non-regular C1 curve. An example: Let y : [0,1] E2 be the curve defined by y(t) = (t3, t3 sin = (0,0) ), 0 < t t = 0. , Let n be some positive integer, and We show that y is not regular. Q : 1 0 < 2 < 4n + 1 < ... < 4n + 1 be a partition of the interval , 1 yt ) - Y(0) [0,1]. 4n + 1 4n + 1 We have 1 ,° ) ( 4n+ 1 , (4n + 1)3 2 Y( 1 1 ) y( ) 4n + 1 4n + 1 = ( 78 ) (4n+1)3 ' (4n + 1)3 7 If 0 is the angle between these two vectors, then cos 0 -, 113 So 0 7.'-± 0 as n . That is, y cannot be smoothly approximated by inscribed polygons. Hence y is not regular. 17 CHAPTER III Measure of Lines That Intersect a Non-Convex Set Measure of Lines that Intersect a Convex Set in E3. 3.1 Let A be a line in E3, and let m(A:An C#(p) be the measure of all lines A that intersect a convex set C in E3. Then (see [2], [3], or [ 4 ]) M(A:An C) =2 IdA S, (3.1) Af1C#(;) Where S is the surface area of C. Consider a line segment OA of length a in E3. Let B(x,r) be a ball B(x,r) is a convex tube. U xcOA The surface area of us denote this tube by T(r,o), or simply by T. of radius r, centered at x. The union Let a T(r, a) Fifyire this tube is 27 ra + 47 r2 1 The measure of all lines A that intersect . the tube T is given by m( A : AfIT# (p) = f dA = (27ra + 4Tr r (3.2) AnT# (1) 3.2. Tube Around Two Intersecting Segments Consider two line segments OA, OB of lengths 01,G2 respectively in E3. Let the angle between them be Y) , 0 < < ff. 18 Let Ti(r, al) = B(x,r), and T2(r, 02) U xE0A = U B(x,r). xEOB Let T = TIU T2. Let A be a line in E3, and let x(A n-r) = number of segments in All T. Then x(All T) = 0, 1, or 2. Figure x(A T) (3.3) 0, if ART = = 2 if A meets T in one segment (that is, if meets T1 fl T2, or, A Ti but not T2, or, T2 but not TO if A meets T in two segments (that is, if A meets both T1 and T2, but not T1 Figures (a), (b) , x(A 11T) = 1, and and (c) rl T2). respectively show the cases where x(AR T) = 0, x(A fl T) = 2. (See following page for figures). Theorem 3.1. f X(AR T) dA A = f dA + T dA An T2 ARTi - I dA A n (Ti fl T2) (3.4) Proof: Case (i): x (ART) = 0 If ART = 4), then All . , An T2, and An (T1 n T2) are all empty. both sides of the equation (3.4) are zero, and the equation is true. Hence 19 Figure 20 x( MIT) = Case (ii): Let AR (T1 A T2) is empty. # 1 of Then neither (1). the sets nn Ti, An T2 So, the lines that intersect T are counted once on the right (Because the lines that intersect T are hand side of equation (3.4). counted in each of the integrals on the right hand side of (3.4), and hence are counted 1 + 1 Let ANT, n T2) = - = 1 So, the equation (3.4) holds. time). 1 Then one of the sets An Ti ,AnT2 is empty and (p. So equation (3.4) holds in this case. the other is not empty and equals ART. Case (iii): Here An (ART) = 2. (T1 n T2) = and A nT2 cp, A (1T1 (1) The lines A that intersect T are counted 1 + 1 - 0 = 2 times on the So the equation holds in this case right hand side of equation (3.4). too. Hence, we have: I (4 X( A 11T) dA = Tr r2+ 27T r02) - 2ff r al) (4r2 + S(Ti n T2), Tr2 Therefore, where S(1-1 A T2) is the surface area of T1 n T2. x (ART) dA 1.3. = 2 (01 2) +4Tr2 r2 2 s(Tin T2). (3.5) Tube Around a Polygon. Consider a polygon lengths uk, k = 1,2,....,n. P in E3, consisting of the edges ek of Let T(r, a,) = U xeek B(x,r). Assume that 21 = (/) for li-j1 li-j1 > 1. To see that this can be done with r small enough, ei Aej = (1) for > Let r be such that T(r, 1. Let min we choose r as follows. Ix =ij xj > 0. n T(r, Let min li-j1 x.ce. G.) = > 1 x.Ee Choose r so that 2r < cS Then . (3.6) T(r, oi) A T(r, = (1) for li-j1 > 1. dge ek of length ak Figure 4 22 Call P closed if el n en is a point such that the last vertex of P in en is also the first vertex of P in e, and ei n ej =0 for 1 < 0 Ii < - ji i - < n - 1. If P is closed we can choose r similarly. n - 1, let Ix. -x min xi E 1 j = 6 ij > 0. ei For 1 Let min 1 < li-j1 < n-1 6 xi = 6. Now choose r so that 2r < 6. T(r, P) =T(r,o1) U and ei Let (3.7) U T(r, 0.11), (All T(r, P)) = number of segments in An T(r, P). (3.8) Theorem 3.2 I x(A n T(r, P)) dA = E k=1 dA An 1-(r, 0k) f - E I dA k=1 An [(T(r, 0k) n T(r, uk4.1)] (3.9) where in the second integral, en+1 stands for el if P is closed, and = en+1 (f) if P is not closed. 23 Proof: Let x(A T(r , P)) = n). The lines A that intersect T(r, P) are counted t times on the left hand side of equation (3.9). By reasoning similar to that in Theorem 3.1, the lines A that intersect T(r, P) are counted 9, times on the right hand side of equation (3.9). Hence we have I x(AnT(r, P)) dA = AI - A2, where n 2 A1 = iT (2 7 rak 471. r2) Tr (3 .10) 2 r LipN + 2 7 2 r2n, k=1 where L(P) = E GI, is the length of P, and k=1 A2 =1L 2 YS[T(r, k=1 ak ) n T(r, akil)] 7 = 2 2 k=1 S(Dk), (3.11) 24 Where 001), Dk = T(r, ok) A T(r, and S(Dk) (3.13) is the surface area of Dk. First let P be closed. Then n 2E S(Dk) A2= (3.14) k=1 SOU + Tr E 2 2 [S(Dk) k=1 k=1 - S(Br) ] 1, Where S(Br) is the surface area of the ball of radius r, in E3. Therefore, 4r2 A2 - [S(Dk) - S(B )] (3.15) k=1 and so, fX(A 72 r L(P) = T(r,P))dA 4 2 [s(1),) - so )] (3.16) k=1 Next, suppose P is not closed. Then n-1 A2 - Tr (21Tr2 + 27r2) + 2 IT2 ok) n E T(r, k=S[T(r, ak+i)] II (the surface area of the two half balls at the ends of P) It n-1 . 2 7. 2 4nr2 + 4ffr2 + 2 7 2 k=1 S(Dk) n-1 { k=1 n-1 s(Br ) + k=, [S(Dk) - S(Br)] 1 25 7 2 IT ffr2 + fx(A [S(D) - 'IT + E 2 n T(r, P))d E 2 n-1 2nr2n = n-1 7 4 Trr2 (n-1) + S(Dk) - S(Br)] k=1 S(Br)], and so k=1 A = n-1 7 Tr2r L(p) E 2 [S(Dk) - S(Br)] (3.18) k=1 we therefore have, 1 fx 2 7r (A nT(r, P)dA - L(P) S(D) - S(Br) 2-7IT n-1 E= where k E if P is closed, and E = if P is not closed. k=1 k=1 Theorem 3.3. S(Dk) - S(Br) 0 asTk 7 , wherek is the angle between the edges ek and ek+, of P. Proof. To show this, let us consider two line segments OA, OB of lengths ol and a2 respectively. Let Y) be the angle AOB. segments draw balls of radius r. T2(r, With centers on these Then we get two tubes Ti(r, al) and 02). Let CI and respectively. C2 be the cylindrical surfaces of the tubes T1 and T2 (That is, CI is the surface of the tube Tl with half balls removed on the two ends of the segment OA). of the surfaces C1 and C2 The intersection is a plane curve, which is in fact an ellipse. 26 ,u 'V I Figure 5 27 Let V be the plane of intersection CI 0 C2. angle between OA and OB. The plane V bisects the In other words, if W is the plane containing the segments OA and OB, the line of intersection of the planes V and W, makes an angle with OA, and an angle +. with OB. Let 0 be the origin, and the axis of C1 (that is, the direction along OA) be the x3-axis, and the plane U through 0 containing end cross section of C1 be the (xl, x2)-plane. Let the line of intersection of the plane W (containing the line segments OA, OB) and the plane U be the x2-axis. Then the line of intersection of the planes U and V will be the x1-axis. Since x3 is perpendicular to the plane U, and OA makes an angle (D F 2 with the line of intersection of the planes V and W, it is clear that the angle between the planes U and V is 7 2 cl Figure 6 2 28 Let us now compute the surface area of the cylinder C1 below If xl is the distance from cl n C2 to the plane (xl, x2), the plane V. then 7 x3 = tan ( 2 cot x2 ) 2 2 x2 - y cot (P 2 sin , where 11) Figure 7 Cross-section of (r, ip) CI is the polar coordinate corresponding to the point (x1 in the plane U. T , x2) Therefore, the surface area of C1 below the plane V 7 = 2 f x3(10ds(fl Since the arc length (1p) = rip, the surface area of the cylinder below the plane V 7 = 2 r2 cot f y-sintp dip 2 = 2r2cot 2 Similarly, the surface area of the cylinder C2 above the plane V = 2r2cot (19 . 2 Hence S(D), the total surface area of T1 n T2, - 2Tr = 4r2cot + 2 = 4r2cot S(D) S(D) 2 2r2(7 +LP). + 2r2(ir +) + = 4r2[cot + 4r[cot + - S(Br) (7O2 '29 -P) 21T - S(Br) = 4r2cot Let F(V) = cot (TT 41rr2 2 - 4nr2 2 TT + (P2 2 1-cosec2 P(v) F(y) 4- F(R) = 0 + But ( ) 2 0 2 as Y" t from 0 to =0 Tr2 0 as IT -> TT Let us now consider a regular space curve y ized by arc length s. Let Q : 0 = so < s1 < : [0,L] E3, parameter- < sn = L be a par- tition of [0,L] and P be the corresponding polygon inscribed in Let ek and s in (0, y. be the kth edge of P obtained by joining the vertices sk-1 By the Approximation Theorem in Chapter IL given . ) E there exists 6 > 0 such that if p(Q) < 6 then lyk - where Yk is the angle between the edges ek and ek+1. Let r < 1 min {ak, k = 1,2,...,n}, where Gk is the length of ek. Trl < 6, 30 Theorem 3 .4. S(Dk) E - 5(Br) 4- 0 as 4- 0. 1_1(Q) Proof. We have, for g2 ' -22 < 4 tan 0 = 0 + Where Bn 1787 285 + + 15 315 + 3 22n(22n - 1) Bn + 82n-1 4..... (2n)! represents the nth Bernoulli number. 7 = tan( cot . 03 2 2 - ) lp + .*. F(Y) = cot Tr+2 ( ' 3 + ( -Y')5 240 24 (7 2 ' )3 7 2 2 2 2 3 cI "s3 for sufficiently small Tr- Co. 4 in (0, Since y is regular, given E if P(Q) < 6 < E. . then - 71 S(Dk) - S(Br) = 4r2 [S(Dk) - S(Br)] Choose r so that r < P('P) there exists 6 > 0 such that Jr -Y013 < r2E3 < nr2c3. ok min < r ) 2 S < min - 5k-1 ,n 31 Then sk - Ski nr < n min z S(Dk) - 2 2 S(B ) L 2 63 ± 0 as p(Q) 0. 32 CHAPTER IV Quermassintegrals of the Intersection of Two Convex Tubes Quermassinte2rals of a Convex Set: convex set, and B be the unit ball, in Eq. B(x,p). X Definition. For p Let C be a 0, let C + pB The set C + pB is called "the parallel set in the C distance p of the convex set C". If V(C + pB) is the volume of C + pB, then we have the formula (see [3] pp. 220-221): q V(C + pB) = p=0 nW (C) pP P P This is called Steiner's formula for parallel convex sets. 0, 1, 2, ... , q, is called the pth quermassintegral of C. W (C), p = The constant term W 0(C) gives the volume of C. The intersection of the convex tubes T1(r') and -12(r a2 ) is a convex set D. In this chapter we show how to compute the quermassintegrals of D. 33 TT + Figure 8 To find the quermassintegrals of D, we compute the volume V(D + pB) We haveD+ pB =DU UB(x,p). xe)D Let us first compute the volume of D. Let x be a ray on the (x1,x2) - plane with xl- axis as the polar axis, 0 as the pole, and let X with the x1-axis. make an angle lp 4 Trr 3 V(D) 3 . Then 7 2 ifixdxdkix3 Tr 34 We have x = X2 3 cot sin - tp cot 2 2 Tr V(D) = 7 1T+ 4Tr r3 2 4r3 3 ( it 7r cot xdxedx3 4)=0 A=o x3=0 + 2 xsi n f + 3 r ) 2 4r3 cm3 --" v, 2 To compute the volume V(D + pB), we di vide the region D + pB into four regions: D, I, II, III so that V(D + pB) = V(D) + V1 + V2+ V3, where V1, V2, V3 are the volumes of the regions I, II, and III, respecti vely. Region I = The region between the balls of radii r and r + p, as shown, Region II = The region between the cylindrical surface C1 of Ti ( r, Gi and the cylindrical surface of T1 ( r + poi) up to the height C1 n C2, and the similar region between the cylindrical surface C2 of T2 ( r, the cyl inarica I surface of T2 ( r + CY 2 ) p, 02), and, Region III = the crescent-like region, above the region II, obtained by adding the balls on C1 n C2. and 35 Figure 9: D + pB / plane x3 = x2 cot 2 /' This portion is extra (that is does not belong to D + pB). 36 The portion of II and III below the plane x x3 = x2 cot line of intersection of the plane x3 = x2 cot and plane (x3, 2 x), a, p sin a) (p cos tip L = tan a= r sin cot Figure 11 = sin ty cot - Je= - r)L 37 Figure 12 x2cot x3 2 Figure 14 Figure 13 iT V2 =41 (I xclxclx3 V3 = 4 f dip ) (I AcIxdx3 ) P2 = ((r + p) cos ty, (r + p) sin ty P3 = + p) cos tp, (r + p ) sin ty P4 = (r cos ty, r sin P5 = ((r + p COS ) chp P3PL,P5 PiPzP3PI, P1 = (r cos q,, r sin ) ) , r sin to cot / r sin tp cot cosg, , (r + p cos a ) sin tP, r sin 4, cot4-+ p sin a) 38 4 Tr (r + p)3 = . 2 (I 0 r dx3 = 4 xdx ) cot-T+ (A-r) tan a r + pCOSa I [ f r sin o r sin 4) cot +p )AdA + dx3 (f p2- + cot r sin IP + p cos a IP f r where tan a = sin d ) o Tr V3 Tr 2 r sin ip.cot r+ p = 4i 2 3 Tr Tr V2 r3 4 rr + Tr 3 (A. - r) 2 dx3 ) Adx] d , cot r sin IP cot -T IV It r+P V2 =4 I r sin IP I cot 2 XdAd r Tr =4 2 f ke cot-T[ r sinIP (r + p) 2 r2 2 ] 05 2 0 = Now (2rp 2r cot + p 2) + (x-r) tan a dx3), AdA + p cos a r sin ip cot = I r + p cos a - AO tan ad?, (2 r sin 4) cot rp 2 cos a sin a 03 COS2 a sin a and 2 3 + p r sin + p cos a dx3 I r sin q) cot ( tt, cot 2_ + ) AdA (x_r) 2 r+p =I /p2 r+p COSa _ (A-0 2 Ack 39 = f p2 _ u2 (r u) du p COS a = 1/2 a - 1/4 sin 2 a) + 1/3 p3 sin 3a rp 2 ( p3 cOS2a V3 sin a 3 -- I 4p3 2 3 sin a dip + f 2rP2 2 I/ 0 4133 3 rp 2 p3 Sirl3aa 2 3 adip sin ip cot 1 92 + 2rp2 dip sin2 + cot2i.) 2 Figure 15 sin ip cot 1 _ 4P3 3 ( 4- ) 2rP2 where TI 2 G(tP) = ) o o 'if dip IT "if = ) I arc tan (sin ip cot Y.' 2 ) dip 2 f arc tan(sinipcot(l°2 )clip 40 V1 = 2/3 V2 = 2r cot V3 - (3r2 (21 (71- (2rp 2122-- 4p3 3 3rP 2 + P + p 2), and 2 G('1, -T-) + 2rp - P3) where Tf 2 GO1 ic arc tan (sin ip cot V(D + pB) = V(D) + [2r2 ( ) dip 71-1-) + 4r2cot + [2r( Tr+ Y)) + 2r cot + 44,o [ 4_1- rr -2- + p + 2r G(Y')] - -2- p2 P3 = Wo(D) + 3 WI(D) + 3 W2(D) + W3(D). W (D) = V(D) - 4r3 4 r3 Tr k-2- 3 W1(D) = 1/3 [2r2(TT+ -2-J + 4r -T- cot 2.- 2 Cot 2-1 = 1/3 S(D), where S(D) is the surface area of D W 2(D) = 1/3 [2r( fl+ + 2r cot -2 + 2r GM], where 7T 2 GCP) = I o W3(D) - . 3 arc tan (sin ip cot -\c- ) dip 41 BIBLIOGRAPHY do Carmo, M., Differential Geometry of Curves and Surfaces, Prentice-Hall, Inc., New Jersey, 1976, pp. 2-22. Kendall, M.G., and Moran, P.A.P., Geometrical Probability, Charles Griffin and Company, Ltd., London, 1963, pp. 16-20, 73-74. Santalo, L.A., Integral Geometry and Geometric Probability, Addison-Wesley Publishing Company, Massachusetts, 1976, pp. 220-227, p. 233. Santa16, L.A., Integral Geometry, Studies in Global Geometry and Analysis. (S.S. Chern, Ed.), Math. Assoc. Amer., 1967, pp. 147-193.