Solutions Manual to Walter Rudin’s Principles of Mathematical Analysis Roger Cooke, University of VermontChapter 10 Integration of Differential Forms Exercise 10.1 Let H be a compact convex set in R* with nonempty interior. Let f € C(H), put f(x) = 0 in the complement of H and define fyy f as in Definition 10.3. Prove that Jy; f is independent of the order in which the integrations are carried out. Hint: Approximate f by functions that are continuous on R* and whose supports are in H, as was done in Example 10.4. Solution. We first give the definition of f,, f, namely Sif, where I is any k-cell containing H. This definition is unambiguous, since if T and J are both k-cells containing H, each of the single integrals carried out is an integral over the same line segment for both cells, namely the intersection of the path of integration with H. There seems to be no way to avoid somehow proving that the boundary of H, denoted OH, has “measure zero.” ‘The definition of the integral as an iterated integral makes that problem slightly more difficult than it would be otherwise, although we can show how to avoid this approach in two dimensions. We shall reserve that discussion until after the proof, which is rather lengthy. The length of the proof is due to the fact that integrals are really defined only over parallelepipeds. The point of the exercise is to enlarge the class of sets over which one can integrate. Our challenge is to show that the boundary of H can be enclosed in a finite set of parallelepipeds whose total volume can be arbitrarily small Our first job is to show that the hypersphere in R has measure zero. AS the proof of that fact involves some work with (k—2)-dimensional hyperspheres in k-dimensional space, we need to make several definitions in order to express these ideas properly. First, for each real number z and each positive number r, S#-?(z) denotes the (k — 2)-dimensional hypersphere in R* having radius r and center at the 175176 CHAPTER 10, 'TEGRATION OF DIFFERENTIAL FORMS point (0,0,...,0,2), that is, Si%(2) = fx: Pt tah ara z. When 2 = 0, we shall write simply $!-? and identify this sphere with the same (k~2)-dimensional hypersphere in R*-1. We observe that S?(z) consists of the two points (r, 2) and (—r,) in R?. Another way of defining the set $#~2(2) is as the intersection of the (k — 1)-dimensional sphere of radius r centered at (0,0,...,0,2) in R* with its equatorial hyperplane P, = {(21,...,24): 2 =z}. Second, for each a € R* and each 6 > 0, IE(5) is the closed hypercube of side 5 in R¥ whose “lower left” corner is a, that is F2(6) = (x: a5 Say Sa 48,5 ey ‘Third, the set of points (m,,...,mx) € RE having integer coordinates will be denoted Z*. Fourth, for all real numbers r and 6 such that 0 < 6 0 and ny = mj +1 if m; < 0, so that (6ny,...,6nx) is a comer of Ism(6). Since [nj| = min(|m;|, lm; + 1)) for all j, (6m) is the unique point of IE, (6) closest to the origin. In particular |6n| < |x|. The lattice point n’ = (ny +e;(m1),...,mg-+ ée(nx)), where €;(t) is 1 if t > 0 and —1 if t < 0, is such that én’ is the corner of Ism(5) opposite to 6n and is the unique point of Kém(6) farthest from the origin. In particular |én’| > |x|. We claim first that jn] — |n| > 1. Indeed, we have InP (nfs tnB) +2(mer(m) es tmaenng)+((enlm)) 2-4-4 (enl))?) (nf +--+) + 2(|na] +--+ + Inel) +2, and (nl +1)? tert) + 2y/nb+ n+l, so that the desired inequality follows from the two inequalities k > 1 and m+ --+nG S Im|-+---+|ng|. This argument shows in general that, for any m € Z*, if b and c are the points in JE,,(6) of minimal and maximal absolute value respectively, then |e| — |b] > 6 From this we deduce a corollary: Let r be any positive real number larger than 8. Ifr <8 r. (For Ten (6) contains the point x of norm s > r. Since Thnn(6) is a connected set, if it contained a point of norm less than or equal tor it would also contain a point of SE) Similarly, if the set TE.,(6) contains no points of ST}, then || < r+ 6. But then it follows that |c| —|b| 1, estimate the number of k-dimensional hypercubes Jf,,(6) that intersect various zones on the (k—1)-sphere S*-? in RE.178 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS We divide the upper hemisphere, consisting of x such that Ix| = rand >0 into half-open zones 2p =A (try... thor, v4) : 22 + wtapar, po 2. We consider all the lattice points m € R¥ such that my = p and If.,(6) intersects Zp. From what we have shown above in Step 1 and Step 2, if m has this property, then the bottom face of I#.,(6) intersects one of the spheres $5; *(p6), where s is an integer such that j < $<. The number of such m is at most N47} and hence is at most 6*-) st~2, which is certainly no larger than. . om" + V@-#) - Because of our estimate of | — j, we see that the total number of m for which mg = p and If,,(6) intersects Zp is at most 2 2p+1 af (Ty? = (1) 2p th _\ fob-2 4 gt-2( (7)? _ 0) wo" ae REE) (4 ((5)-)), VQ) -P where we have used the inequality (1+ £)" < 2% + (2)? with ¢=k-2. We expand this last product into a sum of six terms ate okt geet fy Sea eee geet 8 ( ae (et) ee y omae man f(T)? + t(D? - 2°) Vor +7(Q"-0)*) = 68" (L(p) + (0) + La(P) + Lal) + I5(0) + Io(0)). ‘We need to estimate the sum of each of these terms over p from 0 to [£]180 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS For J(p) we have the simple estimate (§l-1 2 A) = 21] ca) 1 and r > 6, For Io(p) we have (ght reg I 2 he) =24(2) -y a- = = yAEyR-2/7) oe ayy oa SNES) = UE Here we have used the fact that 0 <1— (28) < 1 for the values of p in the range of summation, as we shall do twice more below. For Js(p) we have I; = a x s(p) OS A 6) fF 2 a f dr Oh a 1 = eh [ 4 Dh aR ary fr __y 2) zi = It coba/rybn = 2 582 3) Since Is(p) < 4Js(p) for p = 1,2,...,[§] ~1, it is clear that (gl DAP) < 2? 4 2t-2(Z)t geaZytt, po 6 6 For Is(p) we have Gea >» be) = =O ut i oa <2 (E maha,181 For Je(p) we have get (gla 3 X b= 2G" Y (1 =O Adding all these estimates, we find a sum that is at most (A) (gD ob-1 ok=1y of-2 y gk ok=2y(T)E-] © gk 141 ok—1/T) Rod Ce (<6 ay If we wish to count the total number of hypercubes J#,,(6) that intersect the zones Z,,...,Z;z]-1, we recall our previous observation that such a hypercube can intersect Z, only if my, = p or my = p—1, and if a hypercube If.,(6) with ‘my = p—1 intersects Z, then so does the hypercube Ika.,.,)(6), and the latter has already been counted. Hence in estimating the total number of hypercubes that intersect one of these zones we are more than safe in simply doubling the estimate we have already obtained. As for the zone Zp, any hypercube [f,,(5) with m, = —1 that intersects its bottom edge (the hypersphere St?) also meets the reflection of Zo through the plane xg = 0, and hence the refiection of that hypercube has already been counted among those that meet Zo. When we double our count to include the hypercubes meeting the “southern” hemisphere, all these hypercubes will automatically be counted. Thus it remains only to estimate the hypercubes that meet the “northern arctic zone,” then double the count. Hence we now consider the cap at the top of the hemisphere, whose boundary is the (kt — 2)-dimensional hypersphere st*([516), where 72 6 If mis such that Jf,,(6) meets this set, then we must have (f]-1 4 the coefficient is at most GRUP HIAE/ Deh gi /242 < ght Stage 1 in the proof is now complete. We have shown that the (k—1)-sphere Sto" intersects at most 6" (5)""" hypercubes from the family If.,(8). As the volume of each hypercube is 6*, it follows that the (k — 1)-sphere is contained in a finite union of cubes of total k-dimensional volume 6*'r*-26, Since 6 is an arbitrary positive number, the k-dimensional volume of SE? is zero, ‘We now move on to the second stage of the proof Stage 2. Given any convex set H in R* with non-empty interior, construct a homeomorphism T of R* onto itself that maps S*~! = S*- to aH, the inside of the unit ball to the interior of H, and the outside to the exterior of H, and satisfies a Lipschitz condition on a neighborhood of $*-. To get this result we need some more background work on general convex sets in R* Let C be a bounded convex set in R, and let z be an interior point of C. For each point x on the unit sphere S*? in R¥, let U(x) be the distance from z to the complement of C in the direction of x, that is, U(x) = inf{t > 0: z+tx ¢C}=sup{t > 0:24 tee C}. Step 1. Prove that the function w : S*-1 — (0, +00) is continuous, in fact, that it satisfies a Lipschitz condition: for some constant K,, |yi(x) u(y) < K|x-y| ‘There exist positive numbers a and b such that a < W(x) < b for all x € $*-1 Indeed we can let a be the radius of the largest open ball about z that is contained in @ and b the radius of the smallest closed ball about z containing C. The result now follows from a lemma. Let 0 < 8 < (x). Then C contains the open ball of radius (Q- iw! about atsx. Proof: If s = w(x), this ball is empty, and if s = 0 the assertion is merely the definition of a. Hence assume 0 < 5 < u(x). Now suppose ly — (2+ so] < (1- ra. Let t= 1-2, go that 0 0. Let w = SoG) sa183 s—tr and let u We claim that z+w € C and that 2+ ux € C, so that y=t2+w)+(1-d(@+ux) EC. The first claim will follow if we show that |w| 0 Finally, the last claim is a routine computation: Hatw)+(1-(etux) = ztiw+(1—dux 2+ — (z+ 5x) +trx+t (1—dux = yt(tr—s+(1-t)u)x = y (since (1-t)u= str) ‘The lemma is now proved. Taking y = z+ sv in this lemma (where |v| = 1), we see that y € C (and hence u(v) > s) if wexi<(t ——,—C—“‘ ‘EEC w(x) t such that If ¥(v) >t, we have a fortiori p(v) > t! and = 1 k-vi< Gr ~ aa which, as already shown, implies 14(2x) > 1’, contradicting the choice of t”. There- fore U(v) b~a, the second inequality automatically holds because #(x) and }(v) differ by at most b—a. Specializing to equality in the hypothesis, we deduce the Lipschitz inequality Gx) — v(v)| < bev} We remark that the statement that y belongs to the interior, boundary, and 5 vale ter f lent to [y — 2] < ¥(——), jy - exterior of Cis equivalent to |y — 2] < o(5=4) ly yr2 ~2|>4% a! (y=a) Step 2. Use the function v(x) to define a homeomorphism of R* onto itself that maps S*~1 to OH and is Lipschitz in a neighborhood of S*-!, Such a homeomorphism T(x) is defined for all x € RF as follows, We set T(0) =z and x (x) =24u(% Tix) =2+ “(gs if x 4 0. Since [T(x) — T(0)| < Mjx|, where M = sup{w(y) : ly| = 1}, it is clear that T is continuous at 0. At all other points it is a composition of continuous functions, hence continuous. Since (EIS = X, we have the continuous inverse function xe T(x) -2 © OBC) = 2)/IT) — zl’ That is, for y #z, eo TW)=-> 7, © = Saya which is not 0. Thus the mapping is one-to-one and onto. The mapping also satisfies a Lipschitz condition on the exterior of each ball about 0; that is, on the set Ey = {x : |x| > n} for each 7 > 0. To see this we observe that for any x and y in this set,185 « [o(§) -o(Zpbieo(g Ke es rl +Mix—y| Sl (Be m)ix-vi Here we have used the fact that The statements about the images of the inside of the unit ball, the unit sphere, and the outside are now obvious. For example, as remarked above, y € OW if and only if ly - 2] = “() But this is equivalent to the statement that T-"(y) = pe which says precisely that T-1(y) belongs to the unit sphere. We have now finished Stage 2 of the proof and are ready for the third and final stage. Stage 3. For cach 6 > 0, approximate a function f(x) that is continuous on H by a function fs(x) that is continuous on all of R* and such that the iterated integrals of f and fs differ by at most a fixed multiple of 6 no matter what order they are taken in, To that end, we first let 6 € (0,1/Vk) be given. According to what was proved in Stage 1, the hypersphere $*-1 is contained in the interior of the set of hypercubes 1f,,(6) that intersect it, and there are at most 6¥*61~* of these hypercubes. In each hypercube J#.,(6) from this family we choose and keep fixed one point xm belonging to S*-!. The image of these hypercubes under T is a compact set containing OH in its interior, and each of them is contained in a hypercube of side at most 2L-Vk6 centered at T(xm) € OH, where L is the Lipschitz constant for the mapping TT on the set E15, so that the total volume of these hypercubes is at most 6° (2LV/k)*6. Let c > 0 be the distance from H to the complement of the union of these hypercubes. We define fs(x) as a continuous function that equals f(x) for x € H, while for x not in the interior of HT we set f(x) = max (0,1 - 4%) ¢(6(x)), Here (x) is the unique point of H closest to x and d(x, H) is the distance from x to H. On the boundary of H, where we have apparently given two definitions of fs we have d(x, H) = 0, so that the two definitions are consistent. Hence the piecewise- defined function will be continuous if each of the pieces is, The piece defined186 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS on I is continuous by assumption, so that we need only concern ourselves with the second definition. It is well-known that d(x, H) is a continuous function of x. It is somewhat less obvious that 6(x) is continuous, so that we must. prove that fact. First we show that there is a unique point 6(x) in H closest to x. This is obvious if x € H, so we assume x ¢ H. Let c = min{|x ~ z|: 2 € H}, and suppose 2 and w are two points of H such that |x —z| = ¢ = |x— wl, Then the point w + t(z —w) belongs to H for 0 [x — 2). But [xn ~ 8(Xn)| = (Xn, H) = d(x, H), so that |x —2| = d(x, H) = |x — 4( # contains only one point satisfying this equality, we must have z = 6(x). Thus 6(x) is a continuous function, and therefore f(x) is continuous on all of R¥. It is now clear that |f(x)| and |fs(x)| have the same maximum value, say J and that f and fs differ only on the finite set of hypercubes covering OH. The iterated integrals of the two functions, taken in any order, over this finite set of hypercubes differ by at most 6" JL(2VE)*é. Thus the iterated integral of f differs from the iterated integral of fs by at most this amount, and since all the iterated integrals of fs are equal, it follows that any two iterated integrals of f differ by arbitrarily small amounts, hence are equal. ‘The proof is, at long last, complete. Because this proof is so long and involved, it may be worthwhile to look at an alternative proof that works only for the case k = 2 and does not generalize to higher dimensions. To this end, let k = 2. we define two functions m(-) and M(c), as follows: The domain of both functions is the projection of Hon the a-axis, that is, the set [](H) consisting of x such that there exists y for which (x,y) € H. By definition m(z) is the minimal y for which (x,y) € H, and M(q) is the maximal y for which (x,y) € H. We claim that these functions are continuous on [](#H). Indeed, suppose (2), y(")) € H and x) x. Without loss of generality we can assume that x() > x for all z. (By passing to a subsequence if necessary, we can have either 2) < x for all n or 2(") > x or x(”) = 2 for all n. The last case is trivial, and the other two cases are handled by identical arguments.) Some subsequence of M(x‘) converges to a value 2, Since (x), M(x) € H, and H is closed, it follows that (z, z) belongs to HH. It is clear then that the assumption z > M(z) contradicts the definition of M(z) as the maximal number y for which (x,y) € H. Hence it suffices to prove that z > M(x). This will certainly be the case if M(x) > M{(z) for all n. Hence assume that no is an index for which M(x‘"*)) < M(z). Now 2°) > 2, since if the two were equal, M(z‘"°)) would equal M(x). We observe that if + € [0,1], then the point (tx(°) + (1 - t)x,tM(x')) + (1 — t)M(z)) belongs187 to H. In particular, taking t= 32 =2, we find that to) 4 (1 —d)x = 2), It therefore follows that M(x()) > 20 == y(x(ro) ae Mie) > M(z) ‘Therefore 2 > M(z). (It is this part of the argument that does not generalize to RS, as shown by the the convex set H={(-tty,te): OStS1-1sy 0, we let Hs be the é-neighbothood of H, that is, the set of points whose distance from H is at most 6. Tt is clear that Hs is a convex set containing H in its interior. If f is a continuous function on H, we extend f to a function fg defined on all of R?, as above. By our definition Fesdayae= ff ja,» aya, J ae where [a,0] is the projection of Hon the x-axis and for each x € [a,b] m(z) = min{t : (2,t) € H} and M(z,y) =max(t: (2,t) € H}. We intend to show that the when these integrals are evaluated, the resulting value is the limit of the same integrals evaluated for fs, and of course the same for the integrals in reverse order. Hence these two iterated integrals are equal. To that end, let A be the maximal value of |f(z,y)|, which is also the maxi- mal value of |fs(2,y)|. As we have set f(x,y) = 0 on the complement of H, the two functions f(x,y) and fo(2,) differ only on the set H\\ H, and by no more than A at any point, Let Ps = |a—A(6),b+4(6)] be the projection of Hs on the 2-axis. We claim that A(6) and (6) both tend to zero as 6 tends to zero, For certainly (6) decreases as 6 decreases. Let its limit be c. There is a point (a — (6), ys) © Hs for cach 6 > 0. If y is a limit point of ye as 6 — 0, then, since (a — Xin), 4m) € Hy C He for n < 6 and Hg is closed, it follows that (a—c,y) € He for all 5 > 0, and therefore, since 9, Hs = H, that (a—c,y) € H. By definition of a, it then follows that a < a~¢ —loga(y) if y > 0. Likewise the second sum is finite for each fixed z. The result now follows. The function must be unbounded, since the integral of y; must be 1, even though the support of that function has length 2-*, Exercise 10.3 (a) If F is as in Theorem 10.7, put A = F/(0), Fy(x) = A“1F(x). Then Fi(0) = J. Show that F(x) = Gyo Gy-z 0---0 Gi (x) in some neighborhood of 0, for certain primitive mappings G1,...,@,. This gives another version of Theorem 10.7: F(x) = P'(0)G, 0 G,_1 0--- 0 Gy(x). () Prove that the mapping (x,y) — (y, x) of R? onto R? is not the composition of any two primitive mappings, in any neighborhood of the origin. (This shows that the flips By cannot be omitted from the statement of Theorem 10.7.) Solution: (a) According to the proof of Theorem 10.7, the flips are needed only to interchange m and k, where k is the first index not less than m for which Dmcx(0) #0. Here F%,(0)er YD) (O)e But in that proof F, = F, and since in the present case F’(0) is the identity, By is the identity. But then the definition of G1 (x) as Gilx) implies thet G{ (0) is also the identity. Suppose we know that By, F’,(0), and Gj(0) are all equal to the identity for j < m. Then the inductive definition of Frei a8 Frsi(y) = Fm o Gz(y) implies that Fy, .+1(0) is also the identity, from which it then follows that F’,,,(0), Bm42, and G',,,(0) are all equal to the identity. Thus the decomposition of F; involves no flips, as asserted. x+ [m(x) - Je: (t) If this map were a composition of two primitive maps, its derivative at (0, 0) would be the product of two matrices of the form (a) a= (2 4) second column of the product of the two matrices is zero, which is a contradic- tion. Since this matrix must be a it follows that = 1, d=0. But then the190 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS Exercise 10.4 For (x,y) € R?, define F(e,y) = (€* cosy — 1,¢* siny). Prove that F = G2 0G;, where Gilz,y) = (e* cosy —1,y) Gale.) = (u,(1+u)tanv) are primitive in some neighborhood of (0,0). Compute the Jacobians of Gy, Go, and F at (0,0). Define Ha(z,y) = (2,e* siny) and find Hy (u,v) = (h(u,v),v) so that F = Hj o Hp in some neighborhood of (0,0). Solution: The equation F = Gp o G; is a routine computation, and the fact that G, and G» are primitive is immediate. The Jacobians of G; and G are Vinay . (S700sy —eF siny one 5} 0 Sian (SOY Pr), cana (Qh, ee) so that each of them equals the identity at (0,0). It therefore follows that F’(0,0) = J also, If we take h(u,v) = (ve™—vF ~ 1, ), the primitive mapping Hy (u,v) = (h(u, 0), v) will yield Hy o Hy = F. Exercise 10.5 Formulate and prove an analogue of Theorem 10.8, in which Kis a compact subset of an arbitrary metric space. (Replace the functions 19, that occur in the proof of Theorem 10.8 by functions of the type constructed in Exercise 22 of Chap. 4.) Solution: We are given a compact set K in a metric space X (say with metric 4) and a cover of K by open sets Vi, i=1,2,...,n. (We may as well assume a finite number of sets, since we can find a finite subcover of any infinite cover.) We need to construct continuous functions ¥;, i = 1,2,...,n such that 0 $ vic) < 1 for all i and all z © X, the support of ¢(z) is contained in Vi and > di(x) = 1 for all ze K. = To do this, let 1 > 0 be a Lebesgue number for the covering of K by the sets Vj, that is such that the y-neighborhood of every point « € K is contained in some Vj. Let ¢ € (0,n), and let U; be the set of points whose distance from the complement of V; is larger than ¢ and W; the set of points whose distance191 from the complement of Vj is larger than §. Since the distance from 7 to the complement of V; is a continuous function of z, it follows that U; and W; are open sets. It is obvious that the closure of U; is contained in W; and the closure of W. is contained in Vj. We note that K ¢ 0 Uj. For if x € K there exists Vj such that the n-neighborhood of x is contained in Vj, and hence the distance from « to the complement of that V; is at least 7. Now let A; be the closure of U;, and B; the complement of W;. Define d(x, B;) 0) = Fea) Fale, By Then y;(2) is 1 on A; (and hence certainly on U;) and 0 on Bi, y:(c) is contin- uous, and 0 < g(x) < 1 for all x. Since the support of ¢;(2) is the closure of W,, it is contained in Vi. Since yi(z) > 0 for « in Wi, the sum (2) = 3 ys(2) =f is positive on the open set W = o Wi, which contains K. Now let L be the complement of W, and define a continuous function #(x) by d(z,b) ¥)= TarKy+ ae Dy so that 0 < U(x) <1 for all z, H(z) =1if2€ K, and Y(z)=0if ze L. we now define ¥4(2) = 0 for 2 ¢ W and eilz)v( ez; vile) = then ¥i(z) is continuous on the entire space. Its restriction to L is continuous. If we can show that its restriction to the closure of W’ is continuous, we shall be done. But it is obvious that it is continuous on W itself, and so we need only show that it is continuous at a point of @W. Hence let r, > 2 € OW. Since e:(2)/9(z) is bounded, and Y(z_) — 0, it follows that ¥i(t,) > 0 = ui(z), and hence ¥; is continuous at 2. ‘The construction is now complete. Exercise 10.6 Strengthen the conclusion of Theorem 10.8 by showing that the functions %, can be made differentiable, and even infinitely differentiable. (Use Exercise 1 of Chap. 8 in the construction of the auxiliary functions ¢.) Solution: The function yi(x) is required to have only three properties: 1) (x) = 1 for [x— ail < 755 2) we(x) = 0 for jx —ail > 54; 3) O=< gi(x) < 1 for all x. These properties can be achieved with an infinitely differentiable function (x). To construct such a function, we go to the function f(z) in Exercise 1 of Chapter 8, namely f=e%192 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS for t # 0 and f(0) = 0 = f()(0) for all positive integers n, f()(t) being the nth derivative of f(t). It was established in that exercise that f(t) is infinitely differentiable, and it is obvious that f(t) is strictly increasing for nonnegative values of t. Let 1 t) ott) = £00)= 100) FFM) Then it is obvious that g(t) is an infinitely differentiable function that de- creases from 1 to 0 as x increases from 0 to 1. If we show that (0) = 0 = g)(1) for all positive integers n, it will follow that the function 1 t<0, {#0 Ostsi, 0, 1st is also a C™ function, and we can then take A But it is easy to prove these properties by showing inductively that for all integers j and k with 0 < j 1, it follows that each term vanishes when t = 0 or t= 1, and hence that g(")(1) = 0 = 9()(0) for n = 1,2,... Exercise 10.7 (a) Show that the simplex Q* is the smallest convex subset of Re that contains 0,e1,...,ex. (0) Show that affine mappings take convex sets to convex sets. Solution: (a) By definition Q* = {x : 21 +---m <1, 2; >0,j =1,.. oh}. Tt is obvious that Q* contains all the points 0,e1,...,ex. It is nearly obvious that Q¥ is convex. Indeed, if x and y are points of Q* and 0 <¢ <1, then tx + (1~ fy = 2, where z; = tz; +(1—t)yy. Since 2; > 0 and ys > 0 and O 0, Simple algebra shows that 2 4--.4 2, < t+ (1—1) =1, so that z € Q* also, Thus Q* is convex. Now let C be any convex set containing these points, and let x € Qk We need to show that x € C. We shall show by induction that the point zie + +++ + 2ye; is in C whenever 2 > 0,..., 2; > 0 and +--+ 2; <1, If J = 1, this is obvious, since aye: = aye; + (1 ~ 2,)0 and by assumption O<2 <1 Suppose the theorem is true for j, and let ¢ = 2 +-++4 2541 <1, m > 6,..-.tjs1 2 0. Ife= 0, the point 21¢; +---+2418441 is 0, and hence belongs to C. Therefore we assume c > 0. Since ej41 € C, we need only consider the case j41 <1. By the induetion assumption, taking xj = -2— for l= 1,...,9, we find that the point y = zie, +--+ x4e; belongs to C. and therefore the point (1 — ay41)y + 2y,1€)41 = rey +--+ Tj€y + 2}41€;41 does also. (©) Let A(x) be an affine mapping, that is, A (x) = xp-+T(x), where T(x) is a linear transformation, let C be any convex set, and let u€ A(C), v € A(C). We need to show that tu + (1~t)v € A(C) for all t € (0, 1). But this is trivial, since if u = A(x) and v = A(y), then tu + (1~z)v = A(tx +(1— dy) and ix+(1-dy EC. Exercise 10.8 Let H be the parallelogram in R® whose vertices are (1,1), (3,2), (4,8), (2,4). Pind the affine map T which sends (0,0) to (1, 1), (1,0) to (3,2), (0,1) to (2,4). Show that Jp = 5. Use T to convert the integral a= | eVdedy ie to an integral over J? and thus compute a. Solution: Clearly the constant term in an affine mapping is the image of (0,0), which in the present case is to be (1,1). Thus we are looking for a linear transformation L such that (3,2) = (1,1) + L(1,0) and (2,4) = (1,1) + £(0,1),194 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS which is to say L(1,0) = (2,1) and L(0,1) = (1,3). Obviously L(z,y) = (20 + y,z +3y). Then Jp = 2-3-1 5. The inverse of T is given by T-*(u,v) = L“*((u,v) — (1,1)). Simple algebra then reveals that —2+3u u20) T (uv) = ( 5 , 5 The parallelogram H is the image of the unit square $ under T, and so f eT) Jn | dude. Is Thus Exercise 10.9 Define (x,y) = T(r,6) on the rectangle OSr 0, it follows that the other two are also equal. Exercise 10.10 Let a — oo in Exercise 9, and prove that f,teaaeay= [" [" seorandr, for continuous functions f that decrease sufficiently rapidly as |z| + ly| — oo. (Find a more precise formulation.) Apply this to F(2,y) = exp(—a2? ~ y?) to derive formula (101) of Chap. 8. Solution: Without striving for ultimate generality, we shall assume that there are positive numbers K and 6 such that |f(x,y)] < K(2? + y2)-1-® for all (2,y) # (0,0). (Such an estimate holds for (z,y) ranging over any bounded set merely because f(x,y) is continuous.) Let D, = {(@,y): OS 2? +4? < a} and Sa = {(z,y) : lz] a, so that ala,y) oo 2° A J tesa) ~ halesa)ee Kf” ape Wl WK co s monies de + aay de + [ <[o amet [lames [ae 2Ky|*- ~ 1426 dx + 2K Iyl2=2 < aK y|-1-28, If |y| Sa, we note that f(x,y) = ha(x,y) for —a < x 0, y > 0 integrate wu cover Q, use Theorem 10.9 to convert the integral to one over the strip, and derive formula (96) of Chap. 8 in this way. (For this application, Theorem 10.9 hes to be extended so as to cover certain improper integrals, Provide this extension.) Solution: It is easy to compute the inverse of T, namely saute, ta ~ ere and this inverse is defined on the entire (u,v)-plane with the line v = —u removed. It is obvious that v is positive if and only if s and t have the same sign, and that u is positive if and only if s and 1 —t have the same sign. ‘Thus if u and v are both positive, then t and 1~t have the same sign, which happens if and only if 0 < t <1. In this case s must also be positive. Conversely, the equations that give s and t show that if u and v are both positive, then s > 0 and 0 mao [ eta f stvtem* ds), lo 1/(6=-69/2) ‘The first of these integrals is explicitly calculable and tends to zero as 6 — 0. In the second we use the fact that e~* < 2 for all s > O and taken > n+y+l. It then follows that the integral of f(s — st, st)\ps(s— st, st)s over each of these horizontal line differs from the integral of f(s—st, st)s by an amount that tends to zero uniformly for ¥6 0, Sag <1 (Z* is the unit cube; Q* is the standard simplex in R*.) Define x = T(u) by m= uy 2 (I~ uy)up Te = (L—uy)--+(1— ugar. Show that Show that T maps J* onto Q*, that T is 1-1 in the interior of F*, and that its inverse S is defined in the interior of Q* by u, = 2, and for i= sk. Show that Jr(u) = (2 m)F Hd — ug)? ua), and Js(x) = [A -m)Q= 2 ~2)-Q- ayy). Solution: The first identity is easily proved by induction on k. It is obvious for k=1, and = She k + ote ei (1— uy) (1 ue)ueas +1 = (104) (1 ep) = 1-(1—uy)-+-(1 —ug)(1 — ues). The defining formulas and the formula just proved show that x € Q* when- ever u € J*. In the process of showing that T is onto, we shall prove the inverse ker formula. Let x € Q*, and assume for the moment that D 2 <1. Then all of the equations given as inverse equations are defined. We need only show that the defining equations yield x when applied to the left-hand sides of these equa- tions. Certainly we do have 2, = uy. Suppose that 2, = (1—uy)-++(1—tp-1)uy for r a; = 1, a Ifr = 1, we have xp = +++ = 2, = 0, and this point is its own preimage. In general the preimage of the point x for which 2,41 =--- =z, = 0 is u, where Uy,...,U» are given by the formulas for S. The formulas imply up = 1. The values of tups1,"++;ue are then arbitrary, since the formulas that define T’ will automatically make the remaining x; equal to zero. The Jacobian matrix is a triangular matrix whose diagonal consists of the entries 1, (1— uz), (1- uy)(1—ug),..., (1—uy)--+(1— ta), and this fact yields the formula for Jr(u) immediately. Likewise, the Jacobian of S is triangular and has diagonal entries 1, » Toy==L Sac: from which again the formula for Js(x) is imme- 1 Exercise 10.13 Let r1,...,r; be nonnegative integers, and prove that rilssore! ata de = rel I. 7 « (ktri te +rx)! Hint: Use Exercise 12, Theorems 10.9 and 8.20 Note that the special case r; = --- = r, =0 shows that the volume of Q* is a/R. Solution: Following the hint, we rewrite the integral in terms of u, getting f wD ceufe(L = ay) H CL a gyre. if (2 = ya )*(1 = ur)P8 1 = ta)? (1 — aga) dtr ++ dg. This integral is the product & If a mujhe da, 0202 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS which by formule (96) of Chapter 8 (just proved in Exercise 11 above) equals the product Tyee DRe tt mi ries +r) Dk+2~-ttn trary) When this product is evaluated, the numerator [(k+1— i+ rigs -+-47y) in each factor cancels the denominator [(k+2~(i-+1) +riai ----+r) in the next factor. Thus the product “telescopes” to the product of the factors P(r; + 1) in the numerators divided by the first denominator I(k +1 +r, +--- + ry) Considering that (n+ 1) = n! for integers n, we therefore get the required formula. Theoretically we ought to be worried about the fact that T is not 1-1 on the entire cube J*. This problem, however, is handled by the same reasoning used in Exercises 9, 10, and 11, and need not be repeated. Exercise 10.14 Prove formula (46). Solution: Formula (46) asserts that [] sgn (Jj, — jp) is -1 if the permutation <4 dis-+-sJe i8 odd and 1 if the permutation is even. We observe that this product is (—1)*, where k is the number of pairs (ipsdq) for which jp > jg. Since 880 (Jq ~ Jp) = 1 if Jp < jg and gn (jg — jp) = —1 if jp > Jy, we need to show that the parity of k is the same as the parity of the number of interchanges that will be used in converting this permutation to the identity. (As a corollary, that Parity will be the same, no matter what particular sequence of interchanges is used to get to the identity.) This equality is obvious if the permutation is the identity to begin with. Suppose then that jm > j, and m 2 and o = [po,pi,...,Ps] is an oriented affine k- simplex, prove that 6°7 = 0, directly from the definition of the boundary operator 0. Deduce from this that 0°W = 0 for every chain W. Hint: For orientation, do it first for k = 2, k= 3. In general, if i < j, lot ov be the (k ~ 2)-simplex obtained by deleting p; and p, from o. Show that each aj occurs twice in 0 with opposite sign. Solution. For k = 2 we have 0 = [pr, po} {Po, Po] + {Po, Pi}, so that eo = (po — pr) - (2 ~ po) + (b1 ~ po) For k = 3 we have 90 = [p1,P2+Ps) ~ [Po,P2, Ps] + [Po,P1, Ps] — [Po, Pi, Pa], so that co = (lpe,Ps] ~ [pi,Ps] + (Pi.Pal) ~([P2, Ps] ~ [po, Ps] + (Po, Pe) +((P1,Ps] — [Pos Ps] + [po, P2]) ~([P1, P2] ~ po, P2] + [Po, Ps) In general the order in which p; and p; are omitted from o determines the sign that 04; will have. If p, is omitted first, the resulting (k~ 1)-simplex oj = [Pos --»Pj—1,Pj41;-+-, Px] will acquire the sign (—1)/. If p; is then omitted, the resulting (k — 2)-simplex will acquire a factor of (—1), resulting in (~1)*ox However, if pi is omitted first, pj will move forward one position in the resulting (k ~ 1)-simplex o;, and when it is subsequently omitted, a factor of204 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS (-1))"" will be affixed, resulting in (~1)*3-19,;, Hence the two occurrences of oj; in the second boundary will cancel each other. The linearity of the boundary operator, operating on a base of simplexes, then shows that 0? is the zero operator on all chains. Exercise 10.17 Put J? =7; +7», where 71 = (0,e1,e1 +e2], 72 = —[0,e2,e2 + e1]. Explain why it is reasonable to call J? the positively oriented unit square in R?, Show that 0J? is the sum of 4 oriented affine simplexes. Find these. What is A(t; - 72)? Solution: Although J* is really a collection of two affine mappings, the ranges of these mappings cover the unit square, the diagonal from (0,0) to (1, 1) being covered twice with opposite orientations in the two mappings. In both cases, the sense of orientation is such that the cross product of the last two vertices of the simplex is e3, which is a reasonable definition of the positive orientation on the unit square, By routine computation, oP = (ere +e) —[0,e1 + e2] + (0,e1]) — ([e2,e1 + 2] — [0,e1 + e2] + 0, e2]) = [er,e: + e2] + [ex + 2,2] + [e2, 0] + (0, e). Again, by routine computation, 71-72) = (fer, e: +e2] ~ [0,e1 +e] + [0,e1)) +([e2,e1 + e2] — (0,e1 + e2] + [0, e9]) = [e1,e1 +62] ~ [er + 2,2] — fe2, 0] + [0,e,] — 2[0,e; + en). Exercise 10.18 Consider the oriented affine 3-simplex 01 = [0,e1, €; + €2,e; + €2 + es] in RS. Show that 0; (regarded as a linear transformation) has determinant 1. ‘Thus 01 is positively oriented. Let o2,..., 06 be five other oriented simplexes, obtained as follows: There are five permutations (i, #2, i) of (1,2,3) distinct from (1, 2,3). Associate with each (i1, i2,i3) the simplex S(iis ia, %s)[0,e1,,0:, ey ,e;, +e: +e] where s is the sign that occurs in the definition of the determinant. (This is how 72 was obtained from 7; in Exercise 17.)205 Show that c2,...,06 are positively oriented, Put J° =o, +--+. Then J* may be called the positively oriented unit cube in R® Show that 8J° is the sum of 12 oriented affine 2-simplexes. (‘These 12 triangles cover the surface of the unit cube Js.) Show that x = (¢1,22,2) is in the range of a; if and only if 0 < 23 < 29 < mSl Show that the ranges of o1,...,05 have disjoint interiors, and that their union covers J®, (Compare with Exercise 13; note that 3! = 6.) Solution. We first show that each of these simplexes is positively oriented. To that end, it is convenient to refer to the simplex [0, €:,,€:, + €i.,€i, + €1, + €;5] corresponding to the permutation (i, 2,3) as ofa"), The simplex o*2/s), regarded as a linear transformation, maps (x,y, z) to (2+y+z)e:,+(y+z)ei+ze%,. Its matrix therefore has (1 1 1) asrowé,,(0 1 1) as row i, and (0 0 1) as row is, By interchanging rows in correspondence with the interchanges needed to convert the permutation (i, iz, ig) to the identity, we can convert this matrix to an upper-triangular matrix with 1’s on the main diagonal. The determinant of the matrix is therefore s(iz, i2,i3), so that the simplex s(i1, 42, i3)[0, e:,,€;, + €ig, €:, + €: + €i,] is positively oriented. The boundary of a's) consists of four terms, two of which (lei, ei, + €i2,€;, +€:, + €:,] and —[0,e;,,€;, + ;,]) are not shared with any other o(*) The other two terms (—[0,e;,,¢), +e, + ig] and [0,e:, +€:,,€:, + ei + €] are shared with ot) and g(2:48) respectively. As these two permutations each differ from (i1, 2, ig) by a single interchange, the sign of each of these terms will be opposite in its two occurrences, and hence they will cancel out. Thus the boundary of J* will consist of a total of 12 oriented affine 2-simplexes A point x = (21,2,.23) is in the range of 0} if and only if there are numbers 7,8,t € [0,1] such that r+s+t<1and x = re; +(e; +e2) +#(e: +e2 +e5), that is, 2) = 7r+s+t, 2) = +t, and x5 =t. If such numbers r,s,t exist, obviously 0 < 3 < a2 (-1)'(Boi - Bri), fa in agreement with Exercise 18. Solution. Although we did not spell it out in our solution of Exercise 18, the boundary of J° is the 2-chain DS slevinsia)(lensen + ets.en + eis + eis] — [0,€5,€5 + ei). This sum can be rearranged as a sum of three terms, each of which consists of four terms. For example, the terms in the sum for which ¢.1 can be written as ([erse: + e2,€1 + €2+ 3] . = [er,e: +e3,€1 + €2 + ¢3]) ~ ([0,e,€1 + e2] — [0,e1,€; +e3}) For i; = 2 we get a similar set of four terms, namely, (= [exer + e1,€1 + €2 +3] + [e2,€2 + €3,€; + €2 + €5)) + ([0,€2,¢1 + €2] — [0,e2,€2 + €3]). Finally, for i; = 3 we have ([es,e3 + e1,€1 + €2 + €3] — [es,€3 + €2,€, +e + 5]) ~ ({0,€3,€5 + e1] — 0, e3,e3 +e2]) Now consider the 2-chain fo). According to the notation of Eq. (88), it is Bo(71) + Bor(72). Letting (u,v) = r1(2,y) = (© + y)er + yea, and then (u,v) = 72(2,y) = (e+y)eo+yer (and keeping in mind the orientation assigned to 7%), we see that Jo (z,y) = Bo (t+ y,y) — Bor(y,z + y) = (0.2 +y,y) —207 (0,y,2+4) = [0,e2, e2+es] —[0,e3, €9 +e]. Notice that these two terms occur in the expression for J°, in the groupings for iy = 2 and i, = 3 respectively, but each occurs with the opposite sign. Hence these terms can be accounted fo. in OJ° by being grouped together and written as ~B). Similarly when we look at i, we find that itis the 2-chain whose points are (1,2-+,4) —(1,y,2-+y), which is fer, €: +e2, ¢1 +€2 +e] ~[e1,e1 +e5,€ +e2-+ es]. Again these terms Occur in the expression for e;, this time with exactly the same signs, so that they can be accounted for by grouping them and writing them as the term 1, Thus four of the twelve simplexes in OJ* are accounted for by the expression (-1)*(801 — 811). The other 8 simplexes are accounted for similarly. Exercise 10.20 State conditions under which the formula is valid, and show that it generalizes the formula. for integration by Parts. Hint: d( fo) = (df) Nw + f do. Solution. Given the formula in the hint, we need only invoke Stokes’ Theorem, For any chain satisfying the hypotheses of that theorem we shall have fater= [te which is precisely the given theorem. ‘The ordinary formula for integration by parts follows by considering a 0-forn fg. Exercise 10.21 As in Example 10.36, consider the 1-form _ ty —yde “at eye in R?\ {0}. (a) Carry out the computation that leads to formula (113), and prove that dy = 0. (8) Let () = (rcost,rsint), for some r > 0, and let T be a C’-curve in RY \ {0}, with parameter interval [0,27], with T(0) = [(2n), such that the intervals [7(t),P(¢)] do not contain 0 for any t € [0,2n]. Prove that [o-m : Hint; For 0 0, 6 > 0 are fixed. Use part (b) to show that al ab [ ay at = 2m, lo a? cos? t + Be sin? t 7=a( arctan ®) (d) Show that in any convex open set in which 2 +0, and that z n= a(—arctan ) in any convex open set in which y 4 0. Explain why this justifies the notation 1 = dé, in spite of the fact that n is not exact in R? \ {0}. (e) Show that (b) can be derived from (d) (f) IfT is any closed C’-curve in R? \ {0}, prove that a ala =Ind(X). (See Exercise 23 of Chap. 8 for the definition of the index of a curve.) Solution. (a) By the rules for computing line integrals, given that z = rcost and y =rsint, ae (reost)(r cost) dt ~ (rsint)(—rsint) dt (> fo-f en teen eee eres ea ee =[ dt 1? cos? t +r? sin? t (6) Let P(t) = (X(t), ¥(¢)) and y(t) = (2(t), y(t). Following the hint, observing that the hypothesis that the interval from I(t) to 7(t) does not pass through 0, we find that ®(t,u) is indeed a 2-surface in R? \ {0}, and making it into a singular 2-chain by regarding the domain as an affine 2-chain, as in Exercise 17, we find by Stokes’ theorem that o= [a= 1 e oe =~ foe foe fo- fn209 where 6 is the curve 6(u) = (2, u) = (1—u)P(2x) +u7(2n) and eis the curve e(u) = 8(0,u) = (1-u)P(0) +470). Since 6 and ¢ are the same curve, the last two terms in this expression cancel each other, yielding the required result. (6) We need only verify that ®(t,u) #4 0 = (0,0). But this is clear: Tf (1 — wa + ur) cost = 0, then t= ort 85, since (1—u)a + ur > min(a,r) > 0. But this means that ((1—u)b-+ ur) sint + 0, since ¢ is not a multiple of x. The result now follows. (@) It is @ routine computation that the differential of arctan ¥ is n in the entire right or left half-plane, and similarly for 7 ~ arctan #, which is after all arccot z, Which in tum is arctan ¥ wherever both functions are defined. Thus locally we have 7) = d@, even though 6 is not defined globally in R? \ {0}. (e) Break the integral over + into five parts: 0 < ¢ SP ESt 0. Let (2,y, 2) = O(s,t) define a 2-surface ©, with parameter domain I?, by z=g(t)h(s), y=ag(t)ha(s), 2=g(t)ha(s). Ase 6 directly from (35). Note the shape of the range of &: For fixed s, &(s,t) runs over an interval on a line through 0. The range of ® thus lies in @ “cone” with vertex at the origin, (d) Let E be a closed rectangle in D, with edges parallel to those of D. Suppose J €C"(D), f > 0. Let © be the 2-surface with parameter domain E, defined by Prove that (u,v) = flu,v)D(u,v). Define S as in (b) and prove that f = [c= 415). (Since S is the “radial projection” of © onto the unit sphere, this result makes it reasonable to call eG the “solid angle” subtended by the range of 2 at the origin.) Hint: Consider the 3-surface V given by V(t,u») = [1—t+tf(u,v)]E(u,v), where (u,v) € E,0 W(t,u,v) is a 2-surface ® to which (c) can be applied to show that Jy ¢=0. The same thing holds when w is fixed. By (a) and Stokes’ theorem, Isha211 (e) Put A= ~(z/r)n, where dy —yde ety as in Exercise 21. Then \ isa 1-form in the open set V C R¥ in which 22-44? > 0, Show that ¢ is exact in V by showing that Caan. (f) Derive (d) from (e), without using (c). Hint: To begin with, assume 0 < u <7 on E. By (e), [s-[> and few I, Show that the two integrals of A are equal, by using part (d) of Exercise 21, and by noting that 2/r is the same at D(u,v) as at Q(u, 0). (9) Is ¢ exact in the complement of every line through the origin? Solution. (a) We note that, since 2 = er-}, we have 5(r? — 32). By symmetry we have analogous relations for the partial derivatives of yr~? and 2r~* with respect to y and z respectively. Since dz A dy A dz = dy Adz Adz = dz Ade A dy, we find that de = (7? — 30? +1? — 34? + 7? — 322) de Ady Adz = 0. (6) Since r(E(u,v)) = 1, we have only to note that the differentials pull back to Das dy Adz = $23 duAdv = sin? ucosvduA dv, dz Adz = sin? usinvdurdy and dz Ady = sinucosudu Adv. The integrand then pulls back as (sin? u + sinucos? u) du A dv = sinudu A dv. The reference to Sec. 10.43 must be a misprint for Sec. 10.46. (c) For the application to be made in part (d) below we actually need to al- low the function g(t) to depend on s also. ‘Thus we consider (s,t) instead of o(t). Using only the definition (35) for the integral, we need to get the pullbacks of the wedge products to the parameter domain [0,1] x [0,1]. Since de = 82h; (s) dt + (9(s,t)hi (s) +hy(s)$2 ds, with similar expressions for dy and de, we find that dy A dz = 9(s,t) 32(ha(s)h4(s) — hg(s)ho(s)) ds A dt, dz Ade = 98,4) 32 (N5(s)ha(s) — ho(s)h(s)) ds A dt, and de A dy = 9(,t) 38(h4(s)ho(s) — ‘fa(s)h2(s)) ds Adt. Thus,assuming hy(t), ha(t), and hg(t) do not vanish simul- taneously, we have 1 be : , = _at_{(ha(s)ha(s)ho(s))’ — (Fi (s)ho(s)hg(s)) _ he-Ls He.) Cais) + GaP Gaaye ea O.212 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS (d) Using part (c), as amended, we note that WV consists of six mappings V(1,u,0) = Our), VO,w) = S(uv), V(t,b,0), U(t,a,v), Y(t,u,d), and V(t,u,¢), whete E = [a,)] x c,d]. By part (c) the integrals over each of the last 4 surfaces are all zero. Since d¢ = 0, Stokes’ theorem implies that MoI (e) By straightforward computation, dy = —dz/r)An— (z/r) dq azdr yedys (PP) dz, edz Ady ~ydzAdr ~ FSA es (f) Again by Stokes’ theorem we must have fee ba [ But 17 is independent of z, and 2/r is the same for both S(u,v) and Q(u,») ‘Therefore [ ei an las (9) Yes, ¢ is exact on the complement of every line through the origin. Indeed, for every line through the origin there is a rotation T that maps that line to the z-axis, By Theorem 10.22, part (c) we have d(Ar) = (d\)r = Gr. However, Gr = 6, as one can easily compute. Indeed, since r is invariant under T, we need only show that xdy A dz + ydz A de + zdz A dy is rotation-invariant, To that end, suppose (u,v,w) = T(2,y,2), say u = tye + troy + trax, 80 that du = th de + tedy + tedz, ete, We then have du A dw = (taatey — togtas) dy A dz + (tastsi — taster) da A da + (tortse — taytea) dz A dy, etc. and so udu A dw + vdw A du+wdu A dv works out (after tedious computation) to precisely 2dy Adz +ydzAde+zdz Ady. Exercise 10.23 Fix n. Define ry = (x? +---+23)/? for 1 0, and let wy, be the (k — 1)-form defined in Ey by k we = (ra) SN (H1) hay dry A A dein A digs AoA dzg. a Note that w2 = 1, w3 =¢ in the notation of Exercises 21 and 22. Note also that By C By C++ C Eq = R"\ {0}.213 (a) Prove that day =0 in Ey (8) For k= 2,...,m, prove that wy is exact in Ei-1, by showing that We = d(frwe—1) = (dfx) Awe, where f(x) = (—1)*gx(te/rs) and (1-s)®9ds (-1<¢<1). 1 Hint: f. satisfies the differential equations x: (Vfe)(x) = 0 co (-p* jet = CDE (ren) F- (Defe)(x) = ta (0) Is wy exact in By? (4) Note that (b) is a generalization of part (e) of Exercise 22. ‘Try to extend some of the other assertions of Exercises 21 and 22 to wa, for arbitrary n k Solution. (a) Computation shows that a( Devas dey a---ndesaAdaigs A i “Adzy) kde, A-+- A dey, and $2 = for j < k, so that d(ry) = —k(ru)“*-? D> ay dey, we find that & k diay = (ry) “* dary A Ada — Rr) *? Y abdey As Ady = = Rerny#9 (1B Yo a}) dy A A day =0. jm This argument shows, incidentally, that diz = 0 in E, = R® \ {0}. (0) We compute that ks ~ hry*)dax — J zaeirz® des) it Siam) des = (DMay*(rffdon — rick Yo ziae da). isl ae = (IM -ah rt 9( = (1 rea/r) 8 (re rh a)dre —24 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS Hence, since (dfx) A wy—1 = (—1)*-?we1 A (df), the first term in this last expression contributes kat (re)* (HI) ay ity A A diy Adtigs Noe A dtpaa A dg at to the wedge product. As this contribution is all of wy except the last term ry *(—1)* "lt, dary A---Adz,1, we must endeavor to show that the contribution of the remaining terms amounts to this expression. Since any term containing a repeated factor dr, is zero, we see that the rest of the expression is ba Y8(Yonedas) A (rae)? x cat ka x D(H) ade Ae A dina A dtiga Ao Adena, = (tag (re) “re. which is easily seen to be the same as kaa ge merge, So ahdey A+ dag. (-)* (-1)! rz hay day A+ A dey exactly as required. Thus we have computed this result by “brute force,” arro- gently ignoring the hint. For the benefit of those who wish to use the hint, here is an alternative approach. The wedge product (dfs) Awy-1 is the sum of De fe(x) dz A wy and ke re (So eaDasla)) dor A Adey = CAT? (x (Wa)() ~ teDefi(x)) doy A--Adre, and hence, by the first differential equation, equals Defi) doe N wer — EAT eEDe fil) dary A--- dey, So that the second equation yields the result immediately, The two differential equations themselves are routine computations. (0) No, un is not exact in Ey for any n, since its integral over the (n —1)-sphere pr equals 27z)’ 2S Will be shown below in the answer to part (d). (If it were exact, say the differential of A, this integral would equal thte integral of \ over the boundary of the (n ~ 1)-sphere, which is the 0 (n — 2)-chain.) (d) We can parameterize the (n — 1)-sphere "=! by the mapping T, defined by t= sintsinty---sints.,215 22 = cost; sinty---sinta-1, 2, = costysints---sintn1, Tn-1 = ©08ty—gsintn—1 fy = COStaa1, where 0 costa ae ) Ay sintaa 7 costa 0 0 ~sintai It follows immedietely, when we expand the determinant of the first n — 1 rows along the last column, that (de)... Ada, )y, = sin®~? ta; coStn—1 x oa ata(?-2) a) 1) Ale! ) _ pyrene y(n Oe 2 . _ xX 1) ig at a) dt; Av++Adtyoat a = (UP sin? tar c08taaa (SOA tel P aa? A+ fi Ades) day? A. nas) Adtn1 sin™~? ty 1 008 ty.—1 (sin t sin® tg ---sin"“9 t,_»)dty A-+- A dtn— Hence (HIP (a? def) A Ades) = = (=1)"7? cos? ty sin t sin? tg ---sin"“? ty_y) dty A-++ A dtp. Next, omitting row i (i Indeed, the pullback of wy is ne jam Beh + a8 B+ obs He Ff vere aoe Tagg ges ds, A-++Ad8,_2 dt. [ohn fihn + 98s --- ZB chy gga” Bh | But this determinant is zero, since the first and last columns are proportional. We can now prove that if f(t,...,t.-1) > 0 and bana) = Stas. ytm—r)EMY (ty, teaa)s At,218 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS fen= [on = tAnaalS). To do so, we consider the n-surface in R® given by then V(tas--staoast) = [Ltt tf (ty... te a)BOM (th, staan), for 0 < t < 1 and ty,...,t ranging over a parallelepiped contained in the interior of D with boundary faces parallel to those of D. For each fixed t,, this \W is an (n— 1)-surface of the form just considered, and hence the integral of «, over itis zero. This applies in particular to the faces of the closed parallelepiped EB. Since | wy, = i diy = 0, it then follows that, up to a factor of +1, [uv= [on=a Finally, as in Exercise 22, w, is exact in the complement of every (n — 2)- hyperplane through the origin, since there is a rotation that maps the comple- ment of that hyperplane to Z,—1, while wp is rotation-invariant. a(S). Exercise 10.24 Let w = aa;(x) day be a 1-form of class C” in convex open set EC -R®. Assume du = 0 and prove that w is exact in E by completing the following outline Fix p € E. Define F(x) a w (xe 8). [px] Apply Stokes’ theorem to affine-oriented 2-simplexes [p, x, y] in B. Deduce that a 1 J) - f) = Tux) f ai((1—t)x + ty) dt = forx € E, y € E. Hence D;f(x) = a;(x). Solution. Because dis = 0, the integral of w over the boundary of the oriented 2-simplex [p, x,y] is zero. That is foe [ wtf wno, Soey) Py] (Pex) which can be rewritten as : = Dtu-s) [o-oo at b fy) 400 = [ bey]219 Differentiating with respect to yi, we find : Ditty) = [a(t tiyyat + Ny — 2 ja is tDyay((1—t)x + ty) dt bo The fact that dw = 0 says that Dia; = Dja;, so that we have : Ae Duty) = [ala-ox reyes [Sty -2)Dyed e+ myat = [ota-oxsmnars ['hota—ox+ma S ff esl Ox ty)dt staid ox ty s ai((1—t)x + ty) dt = ay). Thus w = df Exercise 10.25 Assume that w is a 1-form in an open set E C R® such that [ero iy for every closed curve + in E of class C’. Prove that w is exact in Z, by imitating part of the argument sketched in Exercise 24. Solution. We first observe that Stokes’ theorem and the argument of Theorem 10.15 show that dw = 0 in E. (Theorem 10.15 actually shows that if some component of dir is nonzero at some point of H, then there is a 2-surface ® in E whose domain is a 2-cell in R for which f,, dw #0. Then by Stokes’ theorem, fog #0 also, contradicting the assumption of the problem. Tn each connected component Eq of E, we choose a fixed point xq. There is a ball of some positive radius rq centered at Xq and contained in E. Let this ball be Ba. Exercise 24 shows that there is a function (xx) such that w(x) = df(x) inside Ba. By subtracting a constant from f we can assume that f(x) = 0. Now consider the set $ of all points x € E, having the property that there exist a connected open set F containing x and xq and a function fx defined on F, such that df, =w on Fy and fx, = 0. It is clear that $ is an open connected subset of Eq, being the union of all the connected open sets Fx, which have the common point xq. It is also clear that there is a function f defined on $ such that df = w on S. In fact we can define f(x) = fx(x), and this definition is unambiguous, since if f, and fy are both defined at z, then ie)» [tem [ox fo= [aya ne.220 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS Here 7 is a path in Ey from xq to z, and 6 is a path in Ey from xq to z. The path 7 —6 lies in E and is a closed loop, so that L° We need only show that $= E,. But if not, then Eq contains a boundary point x € S. Some open ball B about x is contained in B, and this open ball contains a point y € S. But then there exists a function g such that dg = w in B, and subtracting a constant makes it possible to ensure that gly) = fy(y) = f(y). We claim that g(2) = f(z) on the entire set $B. In fact this argument merely repeats the argument just given to show that f is unambiguously defined. It then follows that y is contained in the connected open set $9 B and that the function h defined to be f on Sand g on B has the property that dh = w on S/N B. By definition, this means y € $, which contradicts the assumption that y is a boundary point of . Therefore $= Ea. Thus we can find a primitive for w on each connected component of E. These primitives can be pieced together to provide a single primitive for w on E. Exercise 10.26 Assume w is a I-form in R° \ {0}, of class C’ and dw = 0 Prove that w is exact in R° \ {0}. Hint: Every closed continuously differentiable curve in R°\ {0} is the bound- ary of a 2-surface in R® \ {0}. Apply Stokes’ theorem and Exercise 25. Solution, Given the assumption in the hint, the solution is easy. By Exercise 25 we need only show that the integral of w over every closed curve is zero. By the assertion in the hint, this closed curve is the boundary of a two-surface. By Stokes’ theorem, the integral of w over the curve equals the integral.of dw over the 2-surface. To prove the claim that every continuously differentiable curve in R® \ {0} is the boundary of a two-surface, we may assume that the curve is of the form x(t), O M2, the projection of the curve x(t) on the unit sphere is contained in a set of spherical caps covering less than half of the upper hemisphere and less than half of the lower hemisphere. Hence there are two antipodal points x9 and —xp on the unit sphere not in its image ‘That means there is at least one line through the origin that the curve does not intersect, This line through the origin gives us a sense of positive rotation from x(t) to x(t+}) for each € (0, 3]. We can then construct a C’-curve 74(s) in R9\ {0} that goes from x(t) to x(t + 3) by letting cylindrical coordinates vary linearly with respect to s. To be specific, we can assume without loss of generality that the line is the 2-axis. In that case, the radial coordinate r(t) = \/2%(t) + °C) is never zero and is a continuously differentiable function of position. We choose 0(t) as the cylindrical polar coordinate of x(t) in a continuously differentiable manner for 0 St < 1, (This is possible by piecing together sections of this function over sufficiently small intervals.) We then define 7(s,¢) = (2(s,t),u(s,¢),2(s,t)) for O a> 0, define (7,8) = (r.cos8,rsin 8) fora v;(y) = 1 for all y € U. Moreover, this sum is locally finite, a that is, each point y has a neighborhood Uy such that the set of indices i for which ¥;(z) # 0 for some z € Uy is finite. To construct such functions, for each x € V, let i(x) be the smallest index r such that x € By. Then, as in the proof of Theorem 10.8, for each x € Ki, choose open balls B(x) and W(x) centered at x such that BUX) C W(x) CWE) C By.224 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS Since K; is compact, there are points x11,...,2¢uy, such that Ky € Bx) U---U Bx). For later convenience we define Ly = Ky Now let Lp = Ky \b Blox). For each x € Lp there are open balls B(x) and W(x) centered at x such that BO) CW(x) C WR) C Bi \ Ki. Since La is compact, we choose a finite set of points x21,...,3aN, such that Lo © B(x) U--U B(x) 2M Notice that Ko. 0, UF (a) De Now suppose we have chosen a (possibly empty) collection of open balls Bins) and W(xus), 1 Ny +--- + Nn41. Therefore the sum and product Yee) =1-J]h- a) jal = are both locally finite at each point. (Local finiteness of the product means all but a finite number of factors equal 1 on a neighborhood of each point.) However, if y € U, then y € B(x;) for some j, and so y;(y) = 1, from which it then follows that Lys) =1 = for all y €U. Since we have defined c; so that x; = (y;,cj), it follows that the projection of the c-section of B(x;) on RP-?, which we denote Cj, is the same as the projection of B(x,) on this subspace. That is, it is B(x;). We can now let aly) = 3 csta(y). For then at each y € U there is a finite integer n such that at (y,a(y)) = daly)(¥se1) +20 + Ual¥)(¥ en). Since U(y) = Of y ¢ Cx and (y.ce) € By C V ify € Cy, it follows that (y,a(y)) is a weighted average of points in V, hence belongs to V for all y € U. Exercise 10.30 If N is the vector given by (135), prove that far Br @28s~as6 det | a2. 2 a3; — af | =|NP. 7 Also, verify Eq. (137). Solution. The equation in the problem is a straightforward computation, and amounts merely to expanding the determinant along the last column. Likewise Eq. (137), which merely asserts that a cross product is perpendicular to each226 CHAPTER 10. INTEGRATION OF DIFFERENTIAL FORMS of the factors, is routine, The two inner products in the equation can be ob- tained by replacing the last column of this determinant by either (a:, a, a3) or (81,82, 83). In each case, the result is a determinant with two equal columns, which is therefore zero. Exercise 10.31 Let EC R® be open, suppose g € C"(E), h € C"(E), and consider the vector field F =9Vh. (a) Prove that V-F = gV7h+ (Vg) -(Vh) where V7h = V-(Vh) = 3) 6%h/82? is the so-called “Laplacian” of h. (b) If Mis a closed subset of E with positively oriented boundary 80 (as in Theorem 10.51), prove that 2, ey 28 fw + (V9) wala = [ot aa where (as is customary) we have written Oh/An in place of (Vh)-n. (Thus Oh/On is the directional derivative of h in the direction of the outward normal to AM, the so-called normal derivative of h.) Interchange g and h, subtract the resulting formula from the first one, to obtain 24 AY2g) dV = oh _ 29 [ov nvtgav = | (0% oF ‘These two formulas are usually called Green’s identities. (c) Assume that h is harmonic in E; this means that V2h = 0. Take g = 1 and conclude that - if FrdA=o Take g = h, and conclude that h = 0 in 9 if h=0 on a0. (d) Show that Green’s identities are also valid in R?. Solution. Part (a) is simply the product rule for derivatives. The main equation in part (b) is simply the divergence theorem applied to F. Green’s identities then follow by completely routine computation. (c) Taking g = 1 forces Og/On = 0 and V2g = 0. Since V2h = 0 by the assumption that h is harmonic, the result follows. For the other assertion of this part we have to go back to the main equation before taking g = h. When we do, we actually get a slightly stronger assertion: Vh = 0 in ©, and so h is constant on each component of ©, if either h = 0 or Oh/On = 0 on all of a0. When h = 0 on 62, obviously the constant value of h must be 0.227 (d) The “two-dimensional” divergence theorem is simply Green's theorem. That is, the assertion that f VF kxF la. fon. follows upon applying Green's theorem to the one-form w = —Fydx + Fydy corresponding to the vector field k x F = —Fhi+F,j. Because the dot and cross operations can be interchanged in the scalar triple product, integrating k x F along a curve, that is, taking the product k x Fr, where r is the tangent to the curve, and then integrating, is the same as integrating F -k x r, which is the normal component of F. All the same identities now follow. Exercise 10.32 Fix 6,0 <6 <1. Let D be the set of all (6,t) € R? such that O0 (1 — 6)? > 0 on all of M, it follows that 7 is defined on M. On T; and P's we have n = 2d0, so that [onw ,