Math 147, Homework 3 Solutions Due: April 24, 2012 1. Let Y ⊂ R3 be the surface of revolution diffeomorphic to S 1 × S 1 you studied in Homework 1 and let f and g be the smooth functions: f, g : Y → R1 defined by f (x, y, z) = x and g(x, y, z) = z. What are the critical values of f ? What are the critical values of g? Sketch the level sets of f and g at their regular values. Solution. Recall Y is a two-dimensional manifold. Since df(x,y,z) = [1, 0, 0], the map df(x,y,z) : T Y(x,y,z) → R1 is surjective unless T Y(x,y,z) = {(x, y, z) | x = 0}. One can check that T Y(x,y,z) = {(x, y, z) | x = 0} happens if and only if (x, y, z) = (−3, 0, 0), (−1, 0, 0), (1, 0, 0), or (3, 0, 0). These are the only critical points of f on Y , and so the only critical values of f are −3, −1, 1, and 3. Since dg(x,y,z) = [0, 0, 1], the map dg(x,y,z) : T Y(x,y,z) → R1 is surjective unless T Y(x,y,z) = {(x, y, z) | z = 0}. One can check that T Y(x,y,z) = {(x, y, z) | z = 0} happens if and only if x2 + y 2 = 2 and z = ±1. These are the only critical points of g on Y , and so the only critical values of g are −1 and 1. A level set of f at a regular value is either a single loop or two adjacent loops. A level set of g at a regular value is a union of two circles, one inside the other. 2. Construct a smooth function f : R1 → R1 whose critical values are dense in R1 . Solution. We will construct a smooth map f : R → R whose set of critical values is the rational numbers Q, which is dense in R. Since the rationals are countable, we can write all rational numbers in a sequence r0 , r1 , r2 , . . .. Using Question 6(d) on Homework 2, for each integer 0 ≤ i we can construct a smooth bump function gi : [i, i + 1] → R1 that is equal to 0 on [i, i+1/5] and on [i+4/5, i+1] and that is equal to ri on [i+2/5, i+3/5]. Function gi has ri as a critical value. Define f : R → R by ( 0 if x < 0 f (x) = gi (x) if i ≤ x < i + 1. Note that f is smooth and has each ri as a critical value. Hence the critical values of f are dense in R1 . 1 3. The unit tangent bundle of a smooth m-manifold M ⊂ Rk is the set of unit vectors in T M : T 1 M = (x, v) : x ∈ M, v ∈ T Mx and |v|2 = 1 . Show that T 1 M is a smooth manifold of dimension 2m − 1. Also show that T 1 M is compact whenever M is compact. Solution. Define f : T M → R by f (x, v) = kvk2 . Then for any (u, w) ∈ T M(x,v) we have df(x,v) (u, w) = 2v · w. So df(x,v) is surjective if v 6= ~0, and so in particular 1 is a regular value of f . Therefore f −1 (1) = T 1 M is a smooth manifold of dimension dim(T M ) − dim(R) = 2m − 1 by Lemma 1 in §2 of Milnor, page 11. Suppose M is compact. Then M is bounded, so there exists some c ∈ R1 with kxk ≤ c for all x ∈ M . Hence for all (x, v) ∈ T 1 M , we have k(x, v)k ≤ kxk + kvk ≤ c + 1. So T 1 M is bounded. Next we show that T 1 M is closed. Suppose (xi , vi ) is a sequence of points in T 1 M converging to some point (x, v) in R2k . We must show that (x, v) ∈ T 1 M . Since M is compact hence closed, we have x ∈ M . Next we’ll show that (x, v) ∈ T M . Consider a local parameterization φ : U → W where W is an open set in X containing x, and where U is an open set in Rk . From page 50 of Guillemin and Pollack, we know dφ : T (U ) → T (W ) is a diffeomorphism. Here T (U ) = U × Rk and dφ acts on a point (u, z) ∈ T (U ) by dφ(u, z) = (φ(u), dφu (z)). Since xi converges to x, for i sufficiently large each xi is in U . And since each vi is in T Mxi , for i sufficiently large each (xi , vi ) is in T (W ). One can show that their preimages (dφ)−1 (xi , vi ) = (ui , zi ) for i sufficiently large lie in some compact subset of T (U ), hence have a subsequence converging to a point (u, z) in T (U ). By continuity of dφ we get a subsequence of (xi , vi ) converging to dφ(u, z). By uniqueness of limit points, we must have (x, v) = dφ(u, z) ∈ T (W ) ⊂ T M. Finally, since k · k is a continuous function and kvi k = 1 for all i, we have kvk = 1, and so (x, v) is also in T 1 M . So T 1 M is closed, bounded, and hence compact. 4. Show that the set SO(3) of 3 × 3 orthogonal matrices of determinant one is diffeomorphic to T 1 S 2 . Solution. 2 Recall T 1 S 2 = {(x, v) : x ∈ S 2 , v ∈ T Sx2 and kvk2 = 1} = {(x, v) ∈ R3 × R3 : kxk = kvk = 1 and x · v = 0}. Define f : T 1 S 2 → SO(3) by letting f (x, v) be the matrix with first column x, second column v, and third column x × v. Map f lands in SO(3) because the resulting matrix has orthonormal columns and has determinant one. Map f is clearly injective. Let’s see why f is surjective. An orthogonal matrix M satisfies M t M = I, and it follows that the first two columns of any orthogonal matrix are orthogonal unit vectors, call them x and v. The third column must be a unit vector orthogonal to the first two, and hence must equal either x × v or −x × v. But for M to also have determinant one, the third column must be x × v. Hence f is surjective. The inverse f −1 is given by sending the first and second columns of a matrix in SO(3) to x and v respectively, forming (x, v) ∈ T 1 S 2 . Since f can be extended to a smooth map F : R3 × R3 → M3 (R) given by the same formula, and since f −1 can be extended to a smooth map G : M3 (R) → R3 × R3 given by the same formula, it follows that f and f −1 are smooth. Hence f : T 1 S 2 → SO(3) is a diffeomorphism. 5. (a) In class we showed that T O(n)I is the set of anti-symmetric matrices in Mn (R). Show that AB − BA is in T O(n)I whenever A and B are in T O(n)I . Solution. Suppose A and B are in T O(n)I and hence anti-symmetric. Then −AT = A and −B T = B. Note that −(AB − BA)T = −B T AT + AT B T = −BA + AB = AB − BA. So AB − BA is also anti-symmetric and hence also in T O(n)I . (b) Let G ⊂ Mn (R) be a collection of matrices that is both a smooth manifold and a group. Show that if A and B are in T GI , then so is AB − BA. Solution. Since A and B are in T GI , from Question 3 on Homework 2 we have smooth maps α, β : (−t, t) → G with α(0) = β(0) = I, dα0 = A, and dβ0 = B. Let g ∈ G. We show that if B ∈ T GI , then the conjugate gBg −1 is also in T GI . Define curve γ : (−t, t) → G by γ(t) = gβ(t)g −1 , which lands in G because G is a group. Since γ(0) = I and γ 0 (0) = gBg −1 , this shows that gBg −1 is in T GI for any g ∈ G. 3 Define curve δ : (−t, t) → T GI by δ(t) = α(t)Bα(t)−1 − B. This curve lands in T GI because α(t)Bα(t)−1 is in T GI for each t by the paragraph above, and since T GI is a linear subspace. Since T GI is a linear subspace, the derivative of curve δ also lies in T GI . Note we have δ 0 (0) = α0 (0)Bα(0)−1 + α(0)B[α(t)−1 ]0t=0 = ABI + IB[α(t)−1 ]0t=0 = AB + B[α(t)−1 ]0t=0 . Since I = α(t)α(t)−1 , we calculate 0 = [I]0t=0 = [α(t)α(t)−1 ]0t=0 = α0 (0)α(0)−1 + α(0)[α(t)−1 ]0t=0 = AI + I[α(t)−1 ]0t=0 = A + [α(t)−1 ]0t=0 . Hence [α(t)−1 ]0t=0 = −A and so δ 0 (0) = AB + B[α(t)−1 ]0t=0 = AB − BA. Hence AB − BA is in T GI . 6. Suppose f : M → N is a smooth map between manifolds, x is in M and dfx is one-to-one. Show that there is a neighborhood of U of x in M such that the restriction of f to U is a diffeomorphism onto its image f (U ). Solution. Let m and n be the dimensions of M and N . Since dfx : T Mx → T Nf (x) is one-to-one, we know that m ≤ n. Moreover, there exists a linear map L : Rn−m → T Nf (x) whose image in T Nf (x) is orthogonal to dfx (T Mx ). Hence (f, L) : M × Rn−m → N is a smooth map whose derivative (dfx , L) : T Mx × Rn−m → T Nf (x) at the point (x, 0) is an isomorphism. By the Inverse Function Theorem on page 13 of Guillemin and Pollack, (f, L) is a local diffeomorphism at (x, 0). That is, there is a neighborhood (U, V ) of (x, 0) in M × Rn−m such that the restriction of (f, L) to (U, V ) is a diffeomorphism onto its image (f (U ), L(V )). Hence the restriction of f to the neighborhood U of x is a diffeomorphism onto its image f (U ). 4 7. Suppose that f : M → N is a smooth map between m-manifolds and that dfx is non-singular at each x in M . Show that f is an open mapping, i.e. f (U ) is open whenever U ⊂ M is open. Solution. Let U ⊂ M be open, and let y be any point in f (U ). Let f (x) = y. Since M and N have the same dimension and dfx is non-singular, dfx is also an isomorphism. By the Inverse Function Theorem on page 13 of Guillemin and Pollack, f is a local diffeomorphism at x. So the restriction of f to some neighborhood U 0 ⊂ U ⊂ M about x is a diffeomorphism onto its image f (U 0 ). Since U 0 is open in M and f is a diffeomorphism here, f (U 0 ) is an open set in N that contains y and that is contained in f (U ). Since y was an arbitrary, we have shown that f (U ) is open in N . 8. (a) Let U be an open set in Rn and f : U → Rn be a smooth map. Show that if A ⊂ U has measure zero, then f (A) has measure zero. [Hint: First show that you can cover A by countably many sets A1 , . . . , An , . . . whose closures Āi are compact and contained in U . Then show that f distorts volumes of cubes by a bounded amount near Āi .] Solution. This is proven in a Theorem in Appendix 1 of Guillemin and Pollack, on the bottom of page 204. How do we write A as a countable union of subsets A1 , A2 , A3 , . . . whose closures are compact and contained in U ? One choice is to let Ai = A ∩ {x ∈ Rn : kxk ≤ i}. (b) Use part (a) to prove mini-Sard’s: If f : U → Rn is a smooth map, with U open in Rm and m < n, then f (U ) has measure zero in Rn . Solution. Define F : U × Rn−m → Rn by F (u, x) = f (u). Map F is smooth because f is smooth. We know that Rm × {0} has measure zero in Rn because m < n, and hence its subset U ×{0} is also a set of measure zero in U ×Rn−m . We apply part (a) to see that F (U × {0}) has measure zero in Rn . Since F (U × {0}) = f (U ), we are done. 5