Quotient Spaces and Quotient Maps Definition. If X is a topological space, Y is a set, and π : X → Y is any surjective map, the quotient topology on Y determined by π is defined by declaring a subset U ⊂ Y is open ⇐⇒ π −1 (U ) is open in X. Definition. If X and Y are topological spaces, a map π : X → Y is called a quotient map if it is surjective and continuous and Y has the quotient topology determined by π. • The fiber of π over a point y ∈ Y is the set π −1 (y). Definition. If π : X → Y is a map, a subset U ⊂ X is said to be saturated with respect to π if U is the entire inverse image of its image U = π −1 (π(U )). • A subset of X is saturated iff it is the union of fibers. Lemma 1 (Properties of Quotient Maps). Let π : X → Y be a quotient map. (a) Characteristic Property: For any topological space B, a map f : Y → B is continuous iff f ◦ π : X → B is continuous. (b) The quotient topology is the unique topology on Y for which the characteristic property holds. Lemma 2 (Passing to the Quotient). Suppose π : X → Y is a quotient map, B is a topological space, and f : X → B is a continuous map that is constant on the fibers of π (i.e., π(p) = π(q) implies f (p) = f (q)). Then there exists a unique continuous map fe : Y → B such that f = fe ◦ π. X πy fe X f y Y −−−−→ B Lemma 3 (Uniqueness of Quotient Spaces). If π1 : X → Y1 and π2 : X → Y2 are quotient maps that are constant on each othere’s fibers (i.e. π1 (p) = π1 (q) iff π2 (p) = π2 (q)), then there exists a unique homeomorphism ϕ : Y1 → Y2 such that ϕ ◦ π1 = π2 . X π1 y ϕ X π y 2 Y1 −−−−→ Y2 . Definition. Let X be a topological space and ∼ is an equivalence relation on X. Let X/ ∼ denote the set of equivalence classes in X, and let π : X → X/ ∼ be the natural projection sending each point to its equivalence class. Endowed with the quotient topology determined by π, the space X/ ∼ is called the quotient space of X determined by the equivalence relation. Typeset by AMS-TEX 1 2 Example (Real Projective Spaces). Definition. The n-dimensional real projective space, denoted by RPn (or sometimes just Pn ), is defined as the set of 1-dimensional linear subspace of Rn+1 . • We give it the quotient topology determined by the natural map π : Rn+1 \ {0} → RPn sending each point x ∈ Rn+1 \ {0} to the subspace spanned by x. — ∀x ∈ Rn+1 \ {0}, denote [x] = π(x) ∈ RPn . • For each i = 1, · · · , n, let ei ⊂ Rn+1 \ {0} = {(x1 , · · · , xn+1 ), xi 6= 0)}, U n e e and let Ui = π(Ui ) ⊂ RP . Since Ui is a saturated open set, Ui is open and ei → Ui is a quotient map. π : U Ũi — Define a map ϕi : Ui → Rn by 1 n+1 ϕi [x , · · · , x ]= xi−1 xi+1 xn+1 x1 . ,··· , i , i ,··· , i xi x x x — This map is well-defined, because its value is unchanged by multiplying x by a nonzero constant. — Because ϕi ◦ π is continuous, ϕi is continuous by the characteristic property of quotient maps. In fact, ϕi is a homeomorphism, because its inverse is given by 1 n 1 i−1 , 1, ui , · · · , un ]. ϕ−1 i (u , · · · , u ) = [u , · · · , u Because the sets Ui cover RPn , this shows that RPn is locally Euclidean of dimension n. RPn is Hausdorff and second countable. RPn is compact, since the restriction of π to Sn is surjective. • The coordinate charts (Ui , ϕi ) constructed in that example are all smoothly compatiable; indeed, for i > j, it is straightforward to compute that 1 uj−1 ui−1 1 ui un u 1 n ϕj ◦ ϕ−1 (u , · · · , u ) = , · · · j , , · · · , j , j , j , · · · , j ), i j u u u u u u which is a diffeomorphism from ϕi (Ui ∩ Uj ) to ϕj (Ui ∩ Uj ). Example (the Complex Projective Space) Definition. Complex projective n-space, denoted by CPn , is defined to be the set of 1-dimensional complex-linear subspaces of Cn+1 , with the quotient topology inherited from the natural projection π : Cn+1 \ {0} → CP n . • CPn+1 is a smooth 2n-manifold. • The quotient map π : Cn+1 \ {0} → CP n is smooth. • CP 1 is diffeomorphic to S2 . 3 Submersions Definition. A smooth map F : M → N of constant rank is called a submersion if F∗ is surjective at each point (or equivalently, if rank F =dim N ). Constant-Rank Level Set Theorem. Let M and N be smooth manifolds, and let Φ : M → N be a smooth map with constant rank to k. Each level set of Φ is a closed embedded submanifold of codimension k in M . Proof. Let c ∈ N be arbitrary, and let S denote the level set Φ−1 (c) ⊂ M . Clearly, S is closed in M by continuity. — To show that is an embedded submanifold, we need to show that ∀p ∈ S, there is a slice chart for S in M near p. — From the rank theorem, there are smooth charts (U, ϕ) centered at p and (V, ψ) centered at c = Φ(p) in which Φ has a coordinate representation of the form Φ(x1 , · · · , xk , xk+1 , · · · , xm ) = (x1 , · · · , xk , 0, · · · , 0), and therefore S ∩ U is the slice {(x1 , · · · , xm ) ∈ U : x1 = · · · xk = 0}. Submersion Level Set Theorem. If Φ : M → N is a submersion, then each level set of Φ is a closed embedded submanifold whose codimension is equal to the dimension of N . Proof. A submersion has constant rank equal to the dimension of N . f → M is any continuous map, a section of π is a continuous Definition. If π : M f M f such that π ◦ σ = IdM : π ↓↑ σ map σ : M → M M. f defined on some open set • A local section is a continuous map σ : U → M U ⊂ M and satisfying the analogous relation π ◦ σ = IdU . Definition. A (smooth) k-dimensional vector bundle is a pair of smooth manifolds E (the total space) and M (the base), together with a surjective (smooth) map π : E → M (the projection), satisfying the following conditions: (1) Each set Ep = π −1 (p) (called the fiber of E over p) is endowed with the structure of a vector space. (2) For each p ∈ M , ∃a nbhd U of p and a homeomorphism (diffeomorphism) ϕ : π −1 (U ) → U × Rk , called a (smooth) local trivilization of E such that the following diagram commutes. ϕ π −1 (U ) −−−−→ U × Rk π1 πy y U U (where π1 is the projection onto the first factor). (3) The restriction of ϕ to each fiber ϕ : Ep → {p}×Rk is a linear isomorphism. 4 Proposition 4 (Properties of Submersion). Let π : M → N be a submersion. Then the following are true. (a) π is an open map. (b) Every point of M is in the image of a smooth local section of π. (b*) π is locally a trivial fibration. (c) If π is surjective, it is a quotient map. Proof. Given p ∈ M , let q = π(p) ∈ N . Because a submersion has constant rank, we can choose smooth coordinates (x1 , · · · , xm ) centered at p and (y 1 , · · · , y k ) centered at q in which π has the coordinte representation π(x1 , · · · , xk , xk+1 , · · · , xm ) = (x1 , · · · , xk ). (b) If ε is sufficiently small positive number, the coordinate cube Cε = {x : |xi | < ε, for i = 1, · · · , m} is a nbhd of p whose image under π is the cube Cε0 = {y : |y i | < ε, for i = 1, · · · , k} The map σ : Cε0 → Cε whose coordinate representation is σ(x1 , · · · , xk ) = (x1 , · · · , xk , 0, · · · , 0) is a smooth local section of π satisfying σ(q) = p. (a) Suppose W is any open subset of M and q = π(W ). – For any p ∈ W with π(p) = q, W contains an open coordinate cube Cε centered at p as above, and thus π(W ) contains an open coordinate cube centered at π(p). – This proves that π(W ) is open, so (a) holds. (c) Since a surjective open map is a quotient map, (c) follows from (a). • The general philosophy of the proofs of the next three propositions is this: To “push” a smooth objects (such as a smooth map) down via a submersion, pull it back via local sections. Proposition 5. Suppose M , N and P are smooth manifolds, π : M → N is a surjectice submersion, and F : N → P is any map. Then F is smooth iff F ◦ π is smooth: M πy M yF ◦π N −−−−→ P. F 5 Proof. (⇒)If F is smooth, then F ◦ π is smooth by composition. (⇐) Suppose that F ◦ π is smooth and let q ∈ N is arbitrary. ∀p ∈ π −1 (q), Proposition 4(b) guarantees the existence of a nbhd U of q M = σ ↑↓ π smooth local section σ : U → M of π such that σ(q) = p. F → U Then π ◦ σ = IdU implies F = F ◦ IdU = F ◦ (π ◦ σ) = (F ◦ π) ◦ σ U U and a M ↓ P . U which is a composition of smooth maps. This shows that F is smooth in a nbhd of each point, so it is smooth. • The next proposition gives a very general sufficient condition under which a smooth map can be “pushed down” by a submersion. Proposition 6 (Passing Smoothly to the quotient). Suppose π : M → N is a surjective submersion. If F : M → P is a smooth map that is constant on the fiber of π, then there is a unique smooth map Fe : N → P such that Fe ◦ π = F : M πy M yF N −−−−→ P. e F Proof. Clearly, if Fe exists, it will have to satisfy Fe(q) = F (p) whenever p ∈ π −1 (q). — We use this to define Fe : Given q ∈ N , let Fe (q) = F (p), where p ∈ M is any point in the fiber over q. (Such a point exists because we are assuming that π is surjective). — This is well-defined because F is constant on the fiber of π, and it satisfies Fe ◦ π = F by construction. — Thus Fe is smooth by Proposition 5. • Our third proposition can be interpreted as a uniqueness result for smooth manifold defined as quotients of other smooth manifolds by submersion. Proposition 7 (Uniqueness of Smooth Quotients). Suppose π1 : M → N1 and π2 : M → N2 are surjective submersions that are constant on each other’s fibers. Then ∃ a unique diffeomorphism F : N1 → N2 such that F ◦ π1 = π2 : M π1 y M π2 y N1 −−−−→ N2 . F 6 Quotients of Manifolds by Group Actions • Suppose a Lie group G acts on a manifold M (on the left, say). — Define a relation on M by setting p ∼ q if there exists g ∈ G such that g · p = q. — This is an equivalence relation, whose equivalence classes are exactly the orbits of G in M . Definition. The set of orbits is denoted by M/G; with the quotient topology it is called the orbit space of the action. • It is of great importance to determine conditions under which the orbit space is a smooth manifiolds. Example. Consider the action of Rk on Rk × Rn by translation in the Rk factor: θv (x, y) = (v + x, y). The orbits are the affine subspaces parallel to Rk , and the orbit space (Rk × Rn )/Rk is homeomorphic to Rn . The quotient map π : Rk × Rn → Rn is a smooth submersion. Lemma 7. For any continuous action of a Lie group G on a manifold M , the quotient map π : M → M/G is open. S Proof. For any open set U ⊂ M , π −1 (π(U )) = g∈G θg (U ). Since θg is a homeomorphism, θg (U ) is open ∀g ∈ U , and therefore π −1 (π(U )) is open in M . Because π is a quotient map, this implies that π(U ) is open in M/G. Theorem 8 (Quotient Manifold Theorem). Suppose a Lie group G acts smoothly, freely, and properly on a smooth manifold M . Then the orbit space M/G is a topological manifold of dimension equal to dimM −dimG, and has a unique smooth structure with the property that the quotient map π : M → M/G is a smooth submersion. Proof. (I) Claim: uniqueness of the smooth structure. — Suppose that M/G has two different smooth structures such that π : M → M/G is a smooth submersion. — Let (M/G)1 and (M/G)2 denote M/G with the first and second smooth structures, respectively. By Proposition 5, the identity map is smooth from (M/G)1 to (M/G)2 : M M π πy y (M/G)1 −−−−→ (M/G)2 Id The same argument shows that it is also smooth in the opposite direction, so the two smooth structures are identical. (II) Prove that M/G is a topological manifold. — Assume for definiteness that G acts on the left, and let θ : G × M → M denote the action and Θ : G × M → M × M the proper map Θ(g, p) = (g · p, p). 7 (i) Claim: M/G is second countable. If {Ui } is a countable basis for the topology of M , then {π(Ui )} is a countable collection of open subsets of M/G (because π is an open map), and it is easy to check that it is a basis for the topology of M/G. (ii) Claim: M/G is Hausdorff. — Define the orbit relation O ⊂ M × M by O = Θ(G × M ) = {(g · p, p) ∈ M × M : p ∈ M, g ∈ G}. (It is called the orbit relation because (g, p) ∈ O iff p and q are in the same G-orbit.) — Since proper continuous maps are closed, O is a closed subset of M × M . — If π(p) 6= π(q), then p and q lie in distinct orbits, so (p.q) ∈ / O. – If U × V is a product nbhd of (p, q) in M × M that is disjoint from O, then π(U ) and π(V ) are disjoint open subsets of M/G containing π(p) and π(q) respectively. (iii) Claim: M/G is locally Euclidean. (iii.a) Claim: The G-orbits are embedded submfds of M diffeomorphic to G. — For any p ∈ M , define the orbit map θ(p) : G → M by θ(p) (g) = g · p. This is a smooth map whose image is exactly the G orbits of p. — Claim: θ(p) is a smooth embedding. (a.1) θp is an injective immersion. (a.1.1) θ(p) is injective. Indeed, if θ(p) (g 0 ) = θ(p) (g), then g 0 · p = g · p, which inplies (g −1 g 0 ) · p = p. Since we are assuming that G acts freely on M , this can happen only if g −1 g 0 = e, which means that g = g 0 . (a.1.2) θ(p) has constant rank. – Indeed, observe that θ(p) (g 0 g) = g 0 · θ(p) (g), θ (p) G −−−−→ Lg0 y M (p) 0 yθ (g ) (θ (p) )∗ Tg G −−−−→ Tθ(p) (g) M (p) 0 (Lg0 )∗ y y(θ (g ))∗ G −−−−→ M θ (p) Tg0 g G −−−−→ Tθp (g0 g) M, (θ (p) )∗ so θ(p) is eqievariant w.r.t. the left translation on G and the given action on M . Since G acts transitively on a G-orbit, the equivariant rank theorem implies that θ(p) has constant rank. (a.2) θ(p) is a proper map. Indeed, if K ⊂ M is compact, then (θ(p) )−1 (K) is closed in G by continuity, and since it is contained in GK∪{p} = {g ∈ G : ∃p ∈ K such that g · p ∈ K ∪ {p}}, it is compact since G acts properly on M . 8 (iii.b) Let k=dim G and n=dim M −dim G. Definition. Smooth chart (U, ϕ) on M , with coordinate functions (x, y) = (x1 , · · · , xk , y 1 , · · · , y n ) is said to be adapted to the G-action if (1) ϕ(U ) is a product open set U1 × U2 ⊂ Rk × Rn , and (2) each orbit intersects U either in the empty set or in a single slice of the form {y 1 = c1 , · · · , y n = cn }. Claim: ∀p ∈ M , ∃an adapted coordinate chart centered at p. (a) To prove this, we begin by choosing any slice chart (W, ϕ0 ) centered at p for the orbit G · p in M . Writing the coordinate functions of ϕ0 as (u1 , · · · , uk , v 1 , · · · , v n ), so that (G · p) ∩ W is the slice {v 1 = · · · = v n = 0}. — Let S be the submfd of W defined by u1 = · · · = uk = 0. (This is the slice “perpndicular” to the orbit in these coordinates.) Thus Tp M decomposes as the following direct sum: Tp M = Tp (G · p) ⊕ Tp S, where Tp (G · p) is the span of (∂/∂ui ) and Tp S is the span of (∂/∂v i ). — Let ψ : G × S → M denote the restriction of the action θ to G × S ⊂ G × M . Claim: ψ∗ : T(e,p) (G × S) → Tp M is surjective, (and then for dimensional reason, it is bijective.) For this, it suffices to claim: Tp (G · p) ⊂ ψ∗ (T(e,p) (G × S)) and Tp S ⊂ ψ∗ (T(e,p) (G × S)). (1) Let ip : G → G × S be the smooth embedding given by ip (g) = (g, p). The orbit map θ(p) : G → M is equal to the composition ip ψ G −−−−→ G × S −−−−→ M. Since θ(p) is a smooth embedding whose image is the orbit G · p, it follows (p) that θ∗ (Te G) is equal to the subapace Tp (G · p) ⊂ Tp M , and thus the image of ψ∗ : T(e,p) (G × S) → Tp M contains Tp (G · p). (2) Similarly, if je : S → G × S is the smooth embedding je (q) = (e, q), then the inclusion ı : S ,→ M is equal to the composition je ψ S −−−−→ G × S −−−−→ M. Therefore, the image of ψ∗ also includes Tp S ⊂ Tp M . — By this and the inverse function theorem, there exist a nbhd X × Y of (e, p) in G × S and a nbhd U of p in M such that ψ : X × Y → M is a diffeomorphism. 9 (b) We need to claim: Y ⊂ S can be chosen small enough that each G-orbit intersects Y in at most a single point. — Suppose this is not the case. Then if {Yi } is a countable nbhd basis for Y at p (e.g., a sequence os coordinate balls whose diameters decrease to 0), for each i there exists distinct points pi , p0i ∈ Yi that are in the same orbit, namely gi · pi = p0i for some gi ∈ G. – Since {Yi } is a nbhd basis, both sequences {pi } and {p0i = gi · pi } converge to p. – Since G acts properly on M , we may pass to a subsequence and assume that gi → g ∈ G. ⇒ g · p = limi→∞ gi · pi = limi→∞ p0i = p, by continuity. ⇒ g = e, since G acts freely. ⇒ gi ∈ X, when i large enough. — But this contradicts the fact that ψ = θ is injective, because X×Y θgi (pi ) = p0i = θe (p0i ). and we are assuming pi 6= p0i . (c) Choose diffeomorphisms α : Bk → X and β : Bn → Y (where Bk and Bn are the open unit balls in Rk and Rn , respectively). Define γ : Bk × Bn → U by γ(x, y) = θα(x) (β(y)). — Because γ is equal to the composition of diffeomorphisms α×β ψ Bk × Bn −−−−→ X × Y −−−−→ U, γ is a diffeomorphism. — The map ϕ = γ −1 is therefore a smooth coordinate map on U . Claim: ϕ is adapted to the G-action. (1) is obvious from the construction. (2) Observe that each y =constant slice is contained in a single orbit, because it is of the form θ(X × {p0 }) ⊂ θ(G × {p0 }) = G · p0 , where p0 ∈ X is the point whose y-coordinate is the given constant. Since an orbit can inersect Y at most once, and each y =constant slice has a point in Y , each orbit intersects U in either precisely one or no slice. 10 (iii.c) To finish the proof that M/G is locally Euclidean, let q = π(p) be an arbitrary point of M/G, and let (U, ϕ) be an adapted coordinate chart for M centered at p, with ϕ(U ) = U1 × U2 ⊂ Rk × Rn . – Let V = π(U ), which is an open subset of M/G because π is an open map. — With the coordinate functions of ϕ denoted by (x1 , · · · , xk , y 1 , · · · , y n ) as before, let Y ⊂ U be the slice {x1 = · · · = xk = 0}. – Note that π : Y → V is bijective by the definition of an adapted chart. – Moreover, if W ia any open subset of Y , then π(W ) = π({(x, y) : (0, y) ∈ W }) is open in M/G, and thus π Y is a homeomorphism. −1 — Let σ = (π Y ) : V → Y ⊂ U , which is a local section of π. Define a map η : V → U2 by sending the equivalence class of points (x, y) to y; this is well-defined by the definition of an adapted chart. — Formally, η = π2 ◦ ϕ ◦ σ, where π2 : U1 × U2 → U2 ⊂ Rn is the projection onto the second factor. σ Y V π η ϕy y U1 × U2 −−−−→ U2 . π2 – Because σ is a homeomorphism from V to Y and π2 ◦ ϕ is a homeomorphism from Y to U2 , it follows that η is a homeomorphism. This completes the proof that M/G is a topological n-manifold. (III) Claim: M/G has a smooth structure such that π is a submersion. — We use the atlas consisting of all charts (V, η) as constructed in the preceeding paragraph. — W.r.t. any such chart from M/G and the corresponding adapted chart for M , π has the coordinate representation π(x, y) = y, which is certainly a submersion. Thus we only need claim: any two such charts for M/G are smoothly compatible. e , ϕ) — Let (U, ϕ) and (U e be two adapted charts for M , and let (V, η) and (Ve , ηe) be the corresponding charts for M/G. (III.1) First consider the case in which the two adapted charts are both centered at the same point p ∈ M . — Write the adapted coordinates as (x, y) and (e x, ye). – The fact that the coordinates are adapted to the G-action means that any two points with the same y-coordinate are in the same orbit, and therefore also have the same ye coordinates. – This means that the transition map between these coordinates can be written (e x, ye) = (A(x, y), B(y)), where A and B are smooth maps defined on some nbhd of the origin. – The transition map ηe ◦ η −1 is ye = B(y), which is clearly smooth. e , ϕ) (III.2) In the general case, suppose (U, ϕ) and (U e are adapted charts for M , and e p ∈ U , pe ∈ U are points such that π(p) = π(e p) = q. 11 — Modifying both charts by adding constant vectors, we can assume that they are centered at p and pe, respectively. — Since p and pe are in the same orbit, ∃a group element g such that g · p = pe. — Because θg is a diffeomorphism taking orbits to orbits, it follows that ϕ e0 = ϕ e ◦ θg is another adapted chart centered at p. – Moreover, σ e0 = θg−1 ◦ σ e is the local section corresponding to ϕ e0 , and therefore e0 ◦ σ e 0 = π2 ◦ ϕ e ◦ θg ◦ θg−1 ◦ σ e = π2 ◦ ϕ e◦σ e = ηe. ηe0 = π2 ◦ ϕ Thus we are back in the situation of the preceeding paragraph, and the two charts are smoothly compatible.