
<p>Prof. A. Cattaneo Institut fu¨r Mathematik Universita¨t Zu¨rich <br>FS 2018 </p><p>Differentiable Manifolds </p><p>Solutions to Exercise Sheet 1 </p><p>√</p><p>Exercise 1 (A non-differentiable manifold). Consider R with the atlas {(R, id), (R, x → sgn(x) x)}. </p><p>Show R with this atlas is a topological manifold but not a differentiable manifold. </p><p>√</p><p>Solution: This follows from the fact that the transition function x → sgn(x) x is a homeomorphism but not differentiable at 0. </p><p>Exercise 2 (Stereographic projection). Let f : S<sup style="top: -0.3012em;">n </sup>− {(0, ..., 0, 1)} → R<sup style="top: -0.3012em;">n </sup>be the stereographic </p><p>projection from N = (0, ..., 0, 1). More precisely, f sends a point p on S<sup style="top: -0.3012em;">n </sup>different from N to the intersection f(p) of the line Np passing through N and p with the equatorial plane x<sup style="top: -0.3013em;">n+1 </sup>= 0, as shown in figure 1. </p><p>Figure 1: Stereographic projection of S<sup style="top: -0.3012em;">2 </sup></p><p>(a) Find an explicit formula for the stereographic projection map f. (b) Find an explicit formula for the inverse stereographic projection map f<sup style="top: -0.3012em;">−1 </sup>(c) If S = −N, U = S<sup style="top: -0.3013em;">n </sup>− N, V = S<sup style="top: -0.3013em;">n </sup>− S and g: S<sup style="top: -0.3013em;">n </sup>→ R<sup style="top: -0.3013em;">n </sup>is the stereographic projection from <br>S, then show that (U, f) and (V, g) form a C<sup style="top: -0.3012em;">∞ </sup>atlas of S<sup style="top: -0.3012em;">n</sup>. </p><p>Solution: We show each point separately. <br>(a) Stereographic projection f : S<sup style="top: -0.3013em;">n </sup>− {(0, ..., 0, 1)} → R<sup style="top: -0.3013em;">n </sup>is given by <br>1</p><p></p><ul style="display: flex;"><li style="flex:1">f(x<sup style="top: -0.3428em;">1</sup>, ..., x<sup style="top: -0.3428em;">n+1</sup>) = </li><li style="flex:1">(x<sup style="top: -0.3427em;">1</sup>, ..., x<sup style="top: -0.3427em;">n</sup>). </li></ul><p></p><p>1 − x<sup style="top: -0.2398em;">n+1 </sup></p><p>(b) The inverse stereographic projection f<sup style="top: -0.3012em;">−1 </sup>is given by <br>1</p><p>2<br>2</p><p></p><ul style="display: flex;"><li style="flex:1">f</li><li style="flex:1"><sup style="top: -0.3428em;">−1</sup>(y<sup style="top: -0.3428em;">1</sup>, ..., y<sup style="top: -0.3428em;">n</sup>) = </li><li style="flex:1">(2y<sup style="top: -0.3428em;">1</sup>, ..., 2y<sup style="top: -0.3428em;">n</sup>, kyk − 1). </li></ul><p></p><p>kyk + 1 </p><p>P</p><p>n</p><p>2</p><p>Here kyk = <sub style="top: 0.249em;">i=1</sub>(y<sup style="top: -0.3013em;">i</sup>)<sup style="top: -0.3013em;">2</sup>. One can then easily check by direct computation that f and the map above are left and right inverses of each other. <br>(c) The transition map g ◦ f<sup style="top: -0.3012em;">−1 </sup>: R<sup style="top: -0.3012em;">n </sup>− {0} → R<sup style="top: -0.3012em;">n </sup>− {0} is given by <br>1</p><p></p><ul style="display: flex;"><li style="flex:1">g ◦ f<sup style="top: -0.3428em;">−1</sup>(y<sup style="top: -0.3428em;">1</sup>, ..., y<sup style="top: -0.3428em;">n</sup>) = </li><li style="flex:1"><sub style="top: 0.3296em;">2 </sub>(y<sup style="top: -0.3428em;">1</sup>, ..., y<sup style="top: -0.3428em;">n</sup>), </li></ul><p></p><p>kyk </p><p>which is smooth, and moreover has a smooth inverse, given by the same formula. The geometric explanation for this is that g ◦ f<sup style="top: -0.3013em;">−1 </sup>represents inversion with respect to the sphere S<sup style="top: -0.3012em;">n−1 </sup>⊂ R<sup style="top: -0.3012em;">n </sup>− {0}, which is an involution (an involution τ of a set X is a map X to X such that τ<sup style="top: -0.3012em;">2 </sup>= id). Hence S<sup style="top: -0.3012em;">n </sup>has a smooth atlas consisting of just two patches (U, f) and (V, g). </p><p>Exercise 3 (Real projective space). Let RP<sup style="top: -0.3012em;">n </sup>= (R<sup style="top: -0.3012em;">n+1 </sup>− {0})/ ∼, where x ∼ tx for all t ∈ </p><p>R − {0}, x ∈ R<sup style="top: -0.3012em;">n+1</sup>, be the n-dimensional real projective space. <br>(a) Prove that RP<sup style="top: -0.3012em;">n </sup>is a differentiable manifold (which is Hausdorff and second countable). </p><p>Solution: We first show that RP<sup style="top: -0.3013em;">n </sup>is Hausdorff and second countable. For this, let π: R<sup style="top: -0.3013em;">n+1 </sup></p><p>→</p><p>RP<sup style="top: -0.3013em;">n </sup>be the canonical projection. </p><p>Claim 1. π is an open map. </p><p>Proof. Let U ⊆ R<sup style="top: -0.3013em;">n+1 </sup>− {0} be open. We have to show that π(U) is open in RP<sup style="top: -0.3013em;">n</sup>. By definition of the quotient topology, we have to show that π<sup style="top: -0.3012em;">−1</sup>(π(U)) is open in R<sup style="top: -0.3012em;">n+1 </sup>− {0}. </p><p>S</p><p>But π<sup style="top: -0.3012em;">−1</sup>(π(U)) = {x ∈ R<sup style="top: -0.3012em;">n+1</sup>, ∃y ∈ U : x ∼ y} = <sub style="top: 0.2491em;">t∈R−{0} </sub>t · U is a union of open sets (since multiplication by non-zero t ∈ R is a homeomorphism of R<sup style="top: -0.3012em;">n+1 </sup>− {0}. Hence π is open. </p><p>Now we get a countable basis for RP<sup style="top: -0.3013em;">n </sup>by taking the image of a countable basis for R<sup style="top: -0.3013em;">n+1 </sup>under π. To show RP<sup style="top: -0.3013em;">n </sup>is Hausdorff, we use the following lemma: </p><p>Lemma 1. Let ∼ be an open equivalence relation on a topological space X(i.e. the canonical projection X → X/ ∼ is open). Then X/ ∼ is Hausdorff if and only if Graph(∼) = {(x, y) ∈ X × X : x ∼ y} is closed. </p><p>P</p><p>To apply the lemma, define f : R<sup style="top: -0.3013em;">n+1 </sup>× R<sup style="top: -0.3013em;">n+1 </sup>→ R by f(x, y) = </p><p><sub style="top: 0.2491em;">i=j</sub>(x<sub style="top: 0.1245em;">i</sub>y<sub style="top: 0.1245em;">j </sub>− x<sub style="top: 0.1245em;">j</sub>y<sub style="top: 0.1245em;">i</sub>)<sup style="top: -0.3013em;">2 </sup>and </p><p>convince yourself that f(x, y) = 0 ⇔ x = ty, t = 0 ⇔ x ∼ y and hence Graph(∼) = f<sup style="top: -0.3013em;">−1</sup>(0) is closed. Therefore RP<sup style="top: -0.3013em;">n </sup>is hausdorff. Next we are going to present a system of charts for RP<sup style="top: -0.3012em;">n</sup>. Let V<sub style="top: 0.1245em;">i </sub>:= {(x<sub style="top: 0.1245em;">1</sub>, . . . , x<sub style="top: 0.1245em;">n+1</sub>) ∈ </p><p>1</p><p>R</p><p><sup style="top: -0.3012em;">n+1 </sup>: x<sub style="top: 0.1246em;">i </sub>= 0}, U<sub style="top: 0.1246em;">i </sub>= π(V<sub style="top: 0.1246em;">i</sub>) and ϕ<sub style="top: 0.1246em;">i </sub>: U<sub style="top: 0.1246em;">i </sub>→ R<sup style="top: -0.3012em;">n</sup>, [(x<sub style="top: 0.1246em;">1</sub>, . . . , x<sub style="top: 0.1246em;">n+1</sub>)] → (x<sub style="top: 0.1245em;">1</sub>, . . . , xˆ , . . . , x<sub style="top: 0.1245em;">n+1</sub>). </p><p>i</p><p>x<sub style="top: 0.083em;">i </sub></p><p>Then one checks that: U<sub style="top: 0.1246em;">i </sub>are open in RP<sup style="top: -0.3013em;">n </sup>(since π is an open map) and their union is RP<sup style="top: -0.3012em;">n</sup>, ϕ<sub style="top: 0.1246em;">i </sub>is well-defined (since ϕ<sub style="top: 0.1246em;">i</sub>([tx]) = ϕ(x)), continuous, and bijective, for all i, with a continuous inverse given by (x<sub style="top: 0.1245em;">1</sub>, . . . , x<sub style="top: 0.1245em;">n</sub>) → [(x<sub style="top: 0.1245em;">1</sub>, . . . , x<sub style="top: 0.1245em;">i−1</sub>, 1, x<sub style="top: 0.1245em;">i</sub>, . . . , x<sub style="top: 0.1245em;">n+1</sub>)] (the continuity follows general topological arguments on quotients). Hence, the ϕ<sub style="top: 0.1245em;">i </sub>are homeomorphisms and {(U<sub style="top: 0.1245em;">i</sub>, ϕ<sub style="top: 0.1245em;">i</sub>)} is an atlas for RP<sup style="top: -0.3013em;">n</sup>. Now we check that the atlas is smooth. For i < j, </p><p>ϕ<sub style="top: 0.1245em;">i </sub>◦ ϕ<sup style="top: -0.3552em;">−</sup><sub style="top: 0.2314em;">j </sub><sup style="top: -0.3552em;">1</sup>(x<sub style="top: 0.1245em;">1</sub>, . . . , x<sub style="top: 0.1245em;">n</sub>) = ϕ<sub style="top: 0.1245em;">i</sub>([(x<sub style="top: 0.1245em;">1</sub>, . . . , x<sub style="top: 0.1245em;">j−1</sub>, 1, . . . , x<sub style="top: 0.1245em;">n</sub>)]) </p><p>1<br>=</p><p>(x<sub style="top: 0.1245em;">1</sub>, . . . , 1, . . . , x<sub style="top: 0.1245em;">n</sub>) </p><p>x<sub style="top: 0.1245em;">i </sub></p><p>is clearly a smooth map (rational with nonzero denominator). Hence RP<sup style="top: -0.3012em;">n </sup>is a smooth manifold. </p><p>(b) Recall that a map between two manifolds f : M → N is smooth if for all charts ϕ<sub style="top: 0.1245em;">α </sub>on M and ψ<sub style="top: 0.1245em;">β </sub>on N, the composition ψ<sub style="top: 0.1245em;">β </sub>◦ f ◦ ϕ<sup style="top: -0.3012em;">−</sup><sub style="top: 0.2052em;">α </sub><sup style="top: -0.3012em;">1 </sup>is smooth in its domain. Also recall that a diffeomorphism is a smooth bijective map with a smooth inverse. Show that RP<sup style="top: -0.3012em;">1</sup>, the real projective line, is diffeomorphic to the circle S<sup style="top: -0.3012em;">1</sup>. </p><p>Solution: Recall that we had a description of S<sup style="top: -0.3012em;">1 </sup>in terms of R/2πZ (example 2.9 in the script). We define a map f : S<sup style="top: -0.3012em;">1 </sup>→ RP<sup style="top: -0.3012em;">1 </sup>by f(θ + 2πZ) = [(cos(θ/2), sin(θ/2)]. </p><p>Claim 2. f is a diffeomorphism </p><p>Proof. First we have to check that f is well-defined. For this, suppose that θ+2πZ = ϕ+2πZ, then θ = ϕ + 2πk for some k ∈ Z and f(ϕ + 2πZ) = [(cos(ϕ/2), sin(ϕ/2)] = [(cos((θ + 2πk)/2), sin((θ + 2πk)/2)] = [(cos(θ/2 + πk), sin(θ/2 + πk)] = [(−1)<sup style="top: -0.3013em;">k</sup>(cos(θ/2), sin(θ/2)] = f(θ + 2πZ). Next, we should check that f is smooth. Recall that on S<sup style="top: -0.3012em;">1</sup>, we can define two charts ψ<sub style="top: 0.1246em;">W </sub>, ψ<sub style="top: 0.1246em;">E </sub>which are the identity maps on different intervals (say (0, 2π) and (−π, π)) and that we have two charts φ<sub style="top: 0.1246em;">1</sub>, φ<sub style="top: 0.1246em;">2 </sub>on RP<sup style="top: -0.3012em;">2 </sup>given by φ<sub style="top: 0.1246em;">1</sub>(x, y) = y/x, φ<sub style="top: 0.1246em;">2</sub>(x, y) = x/y. So, we have to check the 4 representations of f in these charts are smooth: </p><p>f<sub style="top: 0.1246em;">W,U </sub>= φ<sub style="top: 0.1246em;">1 </sub>◦ f ◦ ψ<sub style="top: 0.244em;">W</sub><sup style="top: -0.3552em;">−1 </sup>: (0, 2π) − {π} → R </p><p>1</p><p>θ → φ<sub style="top: 0.1246em;">1</sub>(cos(θ/2), sin(θ/2)) = tan(θ/2) f<sub style="top: 0.1245em;">W,U </sub>= φ<sub style="top: 0.1245em;">2 </sub>◦ f ◦ ψ<sub style="top: 0.244em;">W</sub><sup style="top: -0.3552em;">−1 </sup>: (0, 2π) → R </p><p>2</p><p>θ → φ<sub style="top: 0.1245em;">2</sub>(cos(θ/2), sin(θ/2)) = cot(θ/2) </p><p>f<sub style="top: 0.1245em;">E,U </sub>= φ<sub style="top: 0.1245em;">1 </sub>◦ f ◦ ψ<sub style="top: 0.2439em;">E</sub><sup style="top: -0.3552em;">−1 </sup>: (−π, π) → R </p><p>1</p><p>θ → φ<sub style="top: 0.1245em;">1</sub>(cos(θ/2), sin(θ/2)) = tan(θ/2) </p><p>f<sub style="top: 0.1246em;">W,U </sub>= φ<sub style="top: 0.1246em;">2 </sub>◦ f ◦ ψ<sub style="top: 0.244em;">E</sub><sup style="top: -0.3552em;">−1 </sup>: (0, 2π) − {0} → R </p><p>2</p><p>θ → φ<sub style="top: 0.1246em;">2</sub>(cos(θ/2), sin(θ/2)) = cot(θ/2) where we have to restrict the compositions to ψ<sub style="top: 0.1245em;">A</sub>(f<sup style="top: -0.3012em;">−1</sup>(U<sub style="top: 0.1245em;">i</sub>) for A = E, W. For example, f ◦ ψ<sub style="top: 0.244em;">W</sub><sup style="top: -0.3552em;">−1</sup>(π) = (0, 1) ∈/ U<sub style="top: 0.1245em;">1</sub>. Since all the representations are trigonometric functions, they are smooth. Now we have to find a smooth inverse for f. The correct map is </p><p>(</p><p>2 arctan(y/x) + 2πZ x = 0 g([x, y]) = π + 2πZ x = 0 </p><p>Again, we first have to check the well-definedness of g. This follows from the fact that arctan(ty/tx) = arctan(y/x) and lim<sub style="top: 0.1246em;">t→±∞ </sub>2 arctan(t) = ±π. g is the correct inverse because g◦f(θ+2πZ) = 2 arctan(tan(θ/2)) = θ for θ = π and g◦f(π+ 2πZ) = g(0, 1) = π+2πZ, so g◦f = id<sub style="top: 0.1246em;">S </sub>, and f◦g([x, y]) = [(cos(arctan(y/x)), sin(arctan(y/x))] = </p><p>1<br>1</p><p>[±(x<sup style="top: -0.3012em;">2 </sup>+ y<sup style="top: -0.3012em;">2</sup>)<sup style="top: -0.3012em;">− </sup>(x, y)] = [(x, y)] for x = 0 and f ◦ g([0, y]) = [(0, 1)] = [(0, y)]. As a last step, </p><p>2</p><p>we have to show that g is smooth. One computes </p><p>g<sub style="top: 0.1245em;">U ,W </sub>= ψ<sub style="top: 0.1245em;">W </sub>◦ g ◦ φ<sup style="top: -0.3552em;">−</sup><sub style="top: 0.2213em;">1 </sub><sup style="top: -0.3552em;">1 </sup>: R − {0} → (0, 2π) </p><p>1</p><p>(</p><p>2 arctan(y) y > 0 y → ψ<sub style="top: 0.1246em;">W </sub>(2 arctan(y/1)) = <br>2 arctan(y) + 2π y < 0 </p><p>g<sub style="top: 0.1245em;">U ,W </sub>= ψ<sub style="top: 0.1245em;">W </sub>◦ g ◦ φ<sup style="top: -0.3552em;">−</sup><sub style="top: 0.2213em;">2 </sub><sup style="top: -0.3552em;">1 </sup>: R → (0, 2π) </p><p>2</p><p></p><p>2 arctan(1/x) x > 0 x = 0 π − 2 arctan(1/x) x < 0 </p><p></p><p></p><ul style="display: flex;"><li style="flex:1">x → </li><li style="flex:1">π</li></ul><p></p><p></p><p>g<sub style="top: 0.1245em;">U ,E </sub>= ψ<sub style="top: 0.1245em;">E </sub>◦ g ◦ φ<sup style="top: -0.3552em;">−</sup><sub style="top: 0.2214em;">1 </sub><sup style="top: -0.3552em;">1 </sup>: R → (−π, π) </p><p>1</p><p>y → 2 arctan(y) </p><p>g<sub style="top: 0.1245em;">U ,E </sub>= ψ<sub style="top: 0.1245em;">E </sub>◦ g ◦ φ<sup style="top: -0.3552em;">−</sup><sub style="top: 0.2213em;">2 </sub><sup style="top: -0.3552em;">1 </sup>: R − {0} → (−π, π) </p><p>2</p><p>x → 2 arctan(1/x) <br>The only difficult case is g<sub style="top: 0.1246em;">U ,W </sub>. Here smoothness follows from the fact that all derivatives </p><p>2</p><p>d<sup style="top: -0.2506em;">n </sup></p><p>dx<sup style="top: -0.1661em;">n </sup></p><p>of arctan vanish at infinity, i.e. lim<sub style="top: 0.1245em;">x→±∞ </sub>diffeomorphism. arctan(x) = 0. Hence g is smooth, and f is a <br>Hence RP<sup style="top: -0.3012em;">n </sup>is diffeomorphic to S<sup style="top: -0.3012em;">1</sup>. </p>
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages3 Page
-
File Size-