Chapter 5: Solutions Exercise 5.1 Choose any distinct points p and q on the n-sphere Sn (n > 0). So p and q are unit vectors in Rn+1 . We’ll use a great circle on Sn to connect these points. There are certainly other approaches, e.g., someone from class had the good idea of using longitudes to connect p to q through the “north pole”. The method outlined here works nicely because we won’t need to refer to coordinates for p and q. (a) It may be unlikely, but if it does happens that p ⊥ q (i.e., p · q = 0) then define, for θ ∈ R, the points α(θ) = (cos θ)p + (sin θ)q. 2 2 2 (1) n Since kα(θ)k = α(θ) · α(θ) = cos θ + sin θ = 1 then α(θ) ∈ S for all θ ∈ R. The smooth curve α traces out a great circle on Sn as θ ranges through R with α(0) = p and α(π/2) = q. (b) If q = −p, that is, if p and q are both on a line through the origin (they’re linearly dependent and called antipodal points) then intuitively any great circle containing either p or q should contain the other. Let’s check it: first choose a point e ∈ Sn orthogonal to p and proceed1 as in part (a) above, except with e in place of q: α(θ) = (cos θ)p + (sin θ)e for θ ∈ R. Then α(θ) ∈ Sn for each θ ∈ R and that α(0) = p and α(π) = −p = q. (c) If p and q are not on the same line (i.e., if p and q are linearly independent) then try this approach: replace q with another point e ∈ Sn such that e ⊥ p and e is in the same plane as the points p, q and 0. Then, again as in (a) put α(θ) = (cos θ)p + (sin θ)e. The great circle traced out by α passes through p (θ = 0), e (θ = π/2) and, hopefully, q (θ =?). How do you find the correct θ for q? Well, p · q = cos γ for some angle γ with 0 < γ < π (remember p and q are unit vectors). Setting θ = γ could work. To do this first find the unit vector e and then compute α(γ) to see if α(γ) = q. Construct e by projecting q along p, then subtracting this projection from q, and then normalizing the result to unit length – so that it’s on Sn (this basic construction should be quite familiar to you by now): e= q − (cos γ)p q − (q · p)p q − (cos γ)p q − (q · p)p = =p =p . 2 2 kq − (q · p)pk sin γ 1 − (q · p) 1 − cos γ Now let’s see if α(γ) equals q: q − (cos γ)p α(γ) = cos(γ)p + sin(γ)e = cos(γ)p + sin(γ) = cos(γ)p + q − (cos γ)p = q. sin γ Exercise 5.2 Let p, q ∈ S and α : [a, b] → S a continuous function with α(a) = p and α(b) = q. The composition g ◦ α : I → {−1, 1} is continuous because both α and g are. Suppose g(p) = 1, i.e., (g ◦ α)(a) = 1. If −1 = g(q) = (g ◦ α)(b) then the intermediate value theorem from calculus tells us that, for some t0 ∈ (a, b), (g ◦ α)(t0 ) = 0. Of course, this contradicts the assumption that g assumes only the values −1 and 1. So it must be that g(q) = 1 = g(p). A similar arguments handles the case of g(p) = −1 and we conclude that, in either case, g(q) = g(p) and conclude that either g(q) = 1 for all q ∈ S or g(q) = −1 for all q ∈ S. Exercise 5.4 If f (x1 , . . . , xn+1 ) = x21 + · · · + x2n+1 and if p = (x1 , . . . , xn+1 ) with kpk = r we have ∇f (p) = 2(x1 , . . . , xn+1 ) = 2p with k∇f (p) = 2(x1 , . . . , xn+1 )k = 2r. If N1 (p) = ∇f (p)/k∇f (p)k then N1 (p) = p/r which means, N1 (p) = (p, p/r) and N2 (p) = −N1 (p) = −(p, p/r). Exercise 5.5 (a) n = 1: Rotate N(p) = (p, 0, 1) clockwise 90◦ gives (p, 1, 0). R2 N(p) . ..... ......... .. ... .. ........................... ..... ... ... ... ... .. ... ............................................................ ..... • p R 1 You could construct such a point e from p by first choosing a point r ∈ Sn not on the line through p and q, setting u = r − (r · p)p, and then e = u/kuk. (b) n = 2: In the picture below the two vectors v and Rθ v are in the surface (and tangent space) Sp ' R2p , as is the vector N(p) × v. And I’ve taken θ < 0. The ordered orthogonal basis {v, N(p) × v, N(p)} for R3 is a right-handed system because of the convention we take (in calculus, physics, engineering, etc.) in defining the cross product of two 3-dimensional vectors, in this case, v × [N(p) × v] = kvk2 N(p). R3 N(p) . ...... .......... ... ... ... . ... ................ . . . ... . . ..... . ... . . . ... ........ . ... . . . . . . . . . . . ...... ......... ..... ..... .................................................................... ..... . ............................................... ........................ .... .. .... N(p) × v R2 p• Rθ v θ v (c) n = 3: 1 e1 0 e2 det e3 = det 0 0 N (p) 0 1 0 0 0 0 1 0 0 0 = 1 > 0. 0 1 Exercise 5.6 Follow the text’s hint: Fix p ∈ C and set N (p) = (cos θ, sin θ). Then N(p) = (p, N (p)) = (p, (cos θ, sin θ)) and the positive tangent direction at p is (by definition) the unit vector u = p, cos(θ − π/2), sin(θ − π/2) . Because the tangent space Cp is one dimensional then v = (p, v) must be a scale of this direction, that is, v = ±kvku = ±kvk p, cos(θ − π/2), sin(θ − π/2) . In the figure below v is drawn in the negative tangent direction at p. {v} is inconsistent with orientation N at p. C ... p ..... ..... ..... . . . . ... . . . . ..... .... ..... ..... ..... . . . . . ....... ..... ..... ............ .... .................................................. ..... ........ ..... ........... ..... ....... ................. ..... . . . ...... ..... ......... ..... ..... ..... .... . . . . .. . . . . .... . . . . ....... . . . . ....... ..... .. ..... .. ..... .. .... .... . . . . .... ... ..... .. ..... .. ........... ... ........ . . . . . ... .. ..... ... ..... ... ..... ... ..... ... ... ... ... ... ... ... ..... ... N(p) • p C v The basis {v} for Cp will be consistent with the orientation N on Cp provided v det > 0. N (p) But v cos(θ − π/2) sin(θ − π/2) sin θ det = ±kvk det = ±kvk det N (p) cos θ sin θ cos θ v v So det > 0 if and only if v = kvku, that is, if and only if u = . N (p) kvk 2 − cos θ sin θ = ±kvk. Exercise 5.7 (a) For the first part compute the values of v · (v × w) (and w · (v × w)). You should get 0 (in both cases). To show that kv × wk = kvkkwk sin θ It’s be easier to square both sides first. Here is an identity that may help: 2 v·w v·w If θ ∈ [0, π] is such that cos θ = . This means then sin θ ≥ 0 and sin2 θ = 1 − cos2 θ = 1 − kvkkwk kvkkwk " 2 # v · w kvk2 kwk2 sin2 θ =kvk2 kwk2 1 − kvkkwk =kvk2 kwk2 − (v · w)2 . So it just needs to be checked that: kv × wk2 = kvk2 kwk2 − (v · w)2 . (b) Expand both u · (v × w) and the determinant. The expressions will match. This shows: u u · (v × w) = det v w Swapping different rows in a matrix will change the sign of its determinant. So w v u det v = det w = det u v u w that is, u · (v × w) = v · (w × u) = w · (u × v). (c) If, for any vector u ∈ R3p , u · x = u · (v × w) then u · [x − (v × w)] = 0. Setting u = x − (v × w) gives kx − (v × w)k2 = 0 i.e., x = v × w. Exercise 5.8 (a) By Exercise 5.7(b) and by properties of determinants: N (p) v N(p) · (v × w) = det v = det w > 0 w N (p) if and only if {v, w} is consistent with N(p). (b) Assume, without loss of generality, that both v = (p, v) and w = (p, w) are unit vectors, that is, kvk = kwk = 1. Then v and N(p) × v are perpendicular unit vectors (problem 5.7(a) or because of properties of the cross product you learned in calculus) and so {v, N(p) × w} is an orthonormal basis for the tangent space Sp . Since w is in Sp then there are unique scalars a and b so that w = av + b[N(p) × v]. Take the dot product of w, first with v and then with N(p) × v, to see that a = w · v and b = w · [N(p) × v]. The vectors w and v are on the unit circle in Sp . The image of the rotation Rθ v sweeps through all points on the unit circle exactly once as θ ranges through the angles −π < θ ≤ π. That is, there must be a unique angle θ with −π < θ ≤ π and w = Rθ v. 3 So w = cos θv + sin θ[N(p) × v] for some angle θ with −π < θ ≤ π. This implies that a = cos θ and, more importantly, b = sin θ, so that w v sin θ = b = w · [N(p) × v] = det N (p) = det w . v N (p) v This last expression shows that {v, w} is consistent with the orientation N iff det w > 0 iff sin θ > 0 N (p) iff 0 < θ < π. Exercise 5.9 (a) The vectors v = (p, v), w = (p, w) are given and N(p) = (p, N (p)) depends on the surface S at the point p. u v 4 Regard the expression det w as a function of its first row u = (u1 , u2 , u3 , u4 ) ∈ R . That is, N (p) u v det w = au1 + bu2 + cu3 + du4 = u · x = u · x N (p) where x = (a, b, c, d) ∈ R4 , x = (p, x), u = (p, u) and where the coefficients a, b, c, and d come from the first row determinant expansion and, as such, are uniquely determined by vectors v, w, and N(p) (the other rows). But is x = (p, x) tangent to S at p? We just need to check that x ⊥ N(p). But this is easy now: N (p) v N(p) · x = det w = 0. N (p) So x ∈ Sp and, once the surface S is oriented, depends uniquely on v, w ∈ Sp . Denote by v × w the vector x. (b) Routine computations, many of which follow from properties of determinants. 4