Basic Forms and Orbit Spaces: a Diffeological Approach

his result is a special case of [iglesias, Section 6.38].

ee [IK, Section 2, Paragraph 1].

ore generally, let X be a diffeological space, \sim an equivalence relation on X, and \pi \colon X \to X/\!\!\sim the quotient map. Then the pullback map on forms, \pi^* \colon \Omega^k(X/\!\!\sim) \to \Omega^k(X), is an injection.

By Remark 2.16 it is enough to show that the kernel of the pullback map

is trivial. Let \alpha \in \Omega^k(X/\!\!\sim) be such that \pi^* \alpha = 0. Then, for any plot p \colon U \to X, we have p^* \pi^* \alpha= 0. By the definition of the quotient diffeology, this implies that for any plot q \colon U \to X/\!\!\sim we have q^* \alpha = 0. Hence, \alpha = 0, as required.

o show that \alpha is G-invariant, note that for every g \in G, because \pi\circ g=\pi,

If \alpha is a zero-form (that is, a smooth function) then \alpha is automatically horizontal and we are done. Next, we assume that \alpha is a differential form of positive degree, and we show that \alpha is horizontal. By Lemma 2.21, if x\in M and v\in T_x(G\cdot x), then there exists \xi\in{\mathfrak{g}} such that

Let A_x\colon G\to M be the map sending g to g\cdot x. Then, v=(A_x)_*(\xi).

Thus, to show that \alpha is horizontal, it is enough to show that A_x^*\alpha=0 for all x\in M. Indeed, it follows from the following commutative diagram that A_x^*\alpha=A_x^*\pi^*\beta=0,

ore generally, let X be a diffeological space, let \sim be an equivalence relation on X, and let \pi \colon X \to X/\!\!\sim be the quotient map. We will show that the pullback map

is a diffeomorphism, where the target space is equipped with the subset diffeology induced from \Omega^k(X).

Clearly, \pi^* is surjective to its image. For the injectivity of \pi^*, see Lemma 3.1. As noted in Part (?) of Remark 2.19, \pi^* is smooth. We wish to show that the inverse map

is also smooth.

Fix a plot p \colon U \to \Omega^k(X) with image in \pi^*\Omega^k(X/\!\!\sim). We would like to show that (\pi^*)^{-1} \circ p \colon U \to \Omega^k(X/\!\!\sim) is a plot of \Omega^k(X/\!\!\sim). By Definition 2.18 of the diffeology on spaces of differential forms, we need to show, given any plot r \colon W \to X/\!\!\sim with W \subset {\mathbb{R}}^n, that

is a map to \bigwedge^k {\mathbb{R}}^n that is smooth in (u,w) \in U \times W. Here, (\pi^*)^{-1} is restricted to the image of \pi^*, on which it is well defined because \pi^* is injective.

It is enough to show smoothness locally. For any point w\in W there exist an open neighbourhood V\subseteq W of w and a plot q \colon V\to X such that r|_V = \pi\circ q. For all v\in V, we have

which is smooth in (u,v) \in U \times V by the definition of the standard functional diffeology on \Omega^k(X). And so we are done.

ecause \pi_K is G-equivariant, such a map e exists. Because \pi_G is onto, such a map e is unique. Because the preimage under \pi_K of a G-orbit in N/K is a single G-orbit in N, the map e is one-to-one. Because the maps \pi_{G/K} and \pi_K are onto, the map e is onto.

Thus, the map e is a bijection. To show that it is a diffeomorphism, it remains to show, for every parametrisation p \colon U \to N/G, that p is a plot of N/G if and only if e \circ p is a plot of (N/K)/G.

Fix a parametrisation,

Suppose that p is a plot of N/G. Let u \in U. Then there exist an open neighbourhood W of u in U and a plot q \colon W \to N such that p|_W = \pi_G \circ q. The composition \pi_K \circ q is a plot of N/K, and

Because u \in U was arbitrary, this shows that e \circ p is a plot of (N/K)/G.

Conversely, suppose that e \circ p is a plot of (N/K)/G. Let u \in U. By applying the definition of the quotient diffeology at \pi_{G/K} and then at \pi_K, we obtain an open neighbourhood W of u in U and a plot r \colon W \to N such that

By (?) and by the choice of r,

Because e is one-to-one, this implies that \pi_G \circ r = p|_W. Because u \in U was arbitrary, this shows that p is a plot of N/G. This completes the proof that e is a diffeomorphism.

Because the diagram (?) commutes, \pi_K^* takes \operatorname{Image} \pi_{G/K}^* into \operatorname{Image} \pi_G^*. Because e is a diffeomorphism, we can consider its inverse. From the commuting diagram

we see that \pi_K^* takes \operatorname{Image} \pi_{G/K}^* onto \operatorname{Image} \pi_G^*. Because \pi_K^* is one-to-one (by Lemma 3.1), we have a bijection \pi_K^* \colon \operatorname{Image} \pi_{G/K}^* \to \operatorname{Image} \pi_G^*.

Now suppose that K acts on N properly and with a constant orbit-type, so that N/K is a manifold and \pi_K is a fibre bundle. Then we know that \pi_K^* is a bijection from the differential forms on N/K to the K-basic differential forms on N.

Because \pi_K is G-equivariant, \pi_K^* takes G-invariant forms on N/K to G-invariant forms on N and G-horizontal forms on N/K to G-horizontal forms on N. So we have an injection \pi_K^* \colon \Omega^*_{\rm basic}(N/K) \to \Omega^*_{\rm basic}(N).

Let \alpha be a G-basic form on N. In particular \alpha is K-basic, so there exists a form \beta on N/K such that \alpha = \pi_K^* \beta. Because \pi_K^* is one-to-one and \alpha is G-invariant, \beta is G-invariant. Because \pi_K is G-equivariant and \alpha is G-horizontal, \beta is G-horizontal. This completes the proof that the map \pi_K^* \colon \Omega^*_{\rm basic}(N/K) \to \Omega^*_{\rm basic}(N) is a bijection.

ick a point u \in U and a tangent vector v \in T_uU. Let \xi_1 = (p_1)_*v \left( \in T_{p_1(u)}M \right) and \xi_2 = (p_2)_*v \left( \in T_{p_2(u)}M \right). Let g = a(u). The directional derivative D_v a|_u of a(\cdot) in the direction of v has the form \eta \cdot g \left( \in T_gG \right) for some Lie algebra element \eta (i.e., it is the right translation of \eta by g). We then have that

where g \cdot \xi_1 is the image of \xi_1 under the differential (push-forward) map g_* \colon T_{p_1(u)}M \to T_{p_2(u)}M, and where \eta_M is the vector field on M that corresponds to \eta; in particular \eta_M is everywhere tangent to the G orbits.

Applying this to vectors v^{(1)}, \ldots, v^{(k)} \in T_uU, we get that

his is a consequence of the Baire category theorem.

ix a basic differential k-form \alpha on M. By Proposition 2.20, it is enough to show the following: if p_1 \colon U \to M and p_2 \colon U \to M are plots such that \pi \circ p_1 = \pi \circ p_2, then p_1^* \alpha = p_2^*\alpha on U. Fix two such plots p_1 \colon U \to M and p_2 \colon U \to M. For each g \in G let

By continuity, C_g is closed for each g. By our assumption on p_1 and p_2,

By Lemma 5.3, the set \bigcup_{g \in G} {\operatorname{int}}(C_g) is open and dense in U. Thus, by continuity, it is enough to show that p_1^* \alpha = p_2^*\alpha on {\operatorname{int}}(C_g) for each g \in G. This, in turn, follows from the facts that, for each g\in G, we have g\circ p_1=p_2 on {\operatorname{int}}(C_g) and g^*\alpha=\alpha.

ecause the G_0 action on M is proper, the set

being the image of the proper map (g_0,m) \mapsto (m,g_0 \cdot m), is closed in M \times M.

Fix g' \in \gamma. Because G_0 is normal in G, we can express \gamma as the left coset G_0 g', and we have

We conclude by noting that C_\gamma is the preimage of the closed set \Delta under the continuous map U \to M \times M given by u \mapsto ( g' \cdot p_1(u) , p_2(u) ).

ix a G-basic form \alpha on M. Let \pi \colon M\to M/G be the quotient map, and let p_1 \colon U\to M and p_2 \colon U\to M be plots such that \pi\circ p_1=\pi\circ p_2. Fix \gamma\in G/G_0 and g' \in \gamma, and define C_\gamma as in Lemma 5.5. Define \tilde{p}_1 \colon U \to M as the composition g' \circ p_1. This is a plot of M, and for any u \in {\operatorname{int}}(C_\gamma) we have p_2(u) = g_0 \cdot\tilde{p}_1(u) for some g_0 \in G_0. Consider the restricted action of G_0 on M. Let \pi_0 \colon M \to M/G_0 be the corresponding quotient map. Then the restrictions \tilde{p}_1|_{{\operatorname{int}}(C_\gamma)} and p_2|_{{\operatorname{int}}(C_\gamma)} are plots of M, and they satisfy \pi_0 \circ \tilde{p}_1|_{{\operatorname{int}}(C_\gamma)} = \pi_0 \circ p_2|_{{\operatorname{int}}(C_\gamma)}. By hypothesis, and because \alpha is G_0-basic (as it is G-basic), \alpha is a pullback of a diffeological form on M/G_0. By Proposition 2.20, \tilde{p}_1^* \alpha = p_2^*\alpha on {\operatorname{int}}(C_\gamma).

But on {\operatorname{int}}(C_\gamma) we have \tilde{p}_1^* \alpha = p_1^* {g'}^* \alpha = p_1^* \alpha (since \alpha is G-invariant), and so p_1^* \alpha = p_2^* \alpha on {\operatorname{int}}(C_\gamma). Since \gamma\in G/G_0 is arbitrary, and \bigcup_{\gamma\in G/G_0}{\operatorname{int}}(C_\gamma) is open and dense in U by Lemmas 5.3 and 5.5,

from continuity we have that p_1^*\alpha=p_2^*\alpha on all of U. Finally, by Proposition 2.20, \alpha is the pullback of a form on M/G.

ince V is a vector space, we identify tangent spaces at points of V with V itself. We also identify elements of the group G and of the Lie algebra {\mathfrak{g}} with the matrices by which these elements act on V={\mathbb{R}}^N

So define v':=\int_0^1\left(\exp(t\eta)\cdot v\right)dt. Finally,

This completes the proof.

e claim that there exists a sequence of vectors v'_n converging to 0 in V, and a sequence \mu_n in {\mathfrak{g}}, such that

Indeed, choose any sequence of non-zero real numbers t_n converging to 0. For each n choose \eta_n \in {\mathfrak{g}} such that \exp(\eta_n)=g_{t_n}. Since we are working on a vector space, we can subtract the curves and consider \gamma_2(t) - \gamma_1(t). We have

for some \eta_n \in {\mathfrak{g}} and for some v'_n \in V that satisfy |v'_n| \leq |\gamma_1(t_n)|; the last equality is a result of Lemma 5.7. Because |v_n'| \leq |\gamma_1(t_n)| \underset{n\to\infty}{\to} |\gamma_1(0)| = 0, the claim holds with \mu_n := \eta_n/t_n.

We now have

where (\mu_n)_V is the vector field on V induced by \mu_n\in{\mathfrak{g}}. Because \alpha is horizontal, the last term above vanishes.

et G \times H act on G \times V where G acts by left multiplication on the first factor and where H acts by the anti-diagonal action h \colon (g,v) \mapsto (gh^{-1} , h \cdot v). We have two maps:

Here, the map \pi_H \colon G \times V \to G \times_H V is the quotient by the H-action, and the projection to the second factor {\operatorname{pr}}_2 \colon G \times V \to V can be identified with the quotient by the G-action.

We apply quotient in stages (Lemma 4.1) in two ways: taking the quotient by G and then by H, and taking the quotient by H and then by G. This gives the following commuting diagram:

where e and e' are diffeomorphisms, and \pi_V, \pi_{G\times H}, and \pi are the quotient maps.

By hypothesis, the inclusion \operatorname{Image} \pi_V^* \subset \Omega^*_{\text{basic}}(V) is an equality. Applying Part (iv) of Lemma 4.1 to the left hand side of the diagram (?), we conclude that the inclusion \operatorname{Image} \pi_{G \times H}^* \subset \Omega^*_{\text{basic}}(G \times V) is an equality. Applying Part (iv) of Lemma 4.1 to the right hand side of the diagram (?), we further conclude that the inclusion \operatorname{Image} \pi^* \subset \Omega^*_{\text{basic}} (G \times_H V) is an equality.

y Lemma 5.6, to prove this result for an arbitrary Lie group, it is enough to prove it for the action of the identity component of the group.

For every non-negative integer d, consider the following two statements.

=0pt

1. A(d): For every connected Lie group K with \dim K = d, and for every K-manifold N on which the K-action is proper, every K-basic form on N is the pullback of a differential form on N/K.

2. B(d): For every Lie group K with \dim K = d, and for every K-manifold N on which the identity component of K acts properly, every K-basic form on N is the pullback of a differential form on N/K.

Since the result holds for the trivial group, Statement A(0) is true. By Lemma 5.6, it follows that Statement B(0) is true. Proceeding by induction, we fix a positive integer d, we assume that Statement B(d') is true for all d' < d, and we would like to prove that Statement B(d) is true. By Lemma 5.6, Statement A(d) implies Statement B(d); thus, it is enough to prove that Statement A(d) is true. That is, we may now restrict to the special case of the proposition in which the group is connected and the action is proper, while assuming that the general case of the proposition is true for all Lie groups of smaller dimension.

Now, let G be a connected Lie group, and let M be a G-manifold on which the G-action is proper. Fix a G-basic form \alpha on M. We would like to show that \alpha is the pullback of a differential form on M/G.

By Proposition 2.20 we need to show, for any two plots p_1 \colon W \to M and p_2 \colon W\to M for which \pi \circ p_1 = \pi \circ p_2, that p_1^* \alpha = p_2^* \alpha. Let p_1 and p_2 be two such plots. Fix u\in W. We would like to show that p_1^*\alpha|_u = p_2^*\alpha|_u.

Let x = p_2(u). Let H be the stabiliser of x. By Theorem 2.25 there exists a G-invariant open neighbourhood U of x and an equivariant diffeomorphism F \colon U \to G \times_H V where V = T_xM / T_x(G \cdot x). Because F is an equivariant diffeomorphism and \alpha is G-basic, (F^{-1})^* \alpha is G-basic on G \times_H V.

Either x is a fixed point, or x is not a fixed point.

Suppose that x is a fixed point. Then H=G. So p_1(u) = p_2(u) = x, and F identifies U with V = T_xM, sending x to 0 \in V. Fixing a G-invariant Riemannian metric, we have that G acts linearly and orthogonally on V. Let v \in T_uW. Applying Lemma 5.8 to the curves \gamma_1(t) := F(p_1(u+tv)) and \gamma_2(t) := F(p_2(u+tv)) in V and to the basic form (F^{-1})^*\alpha on V, we obtain that \dot{\gamma}_1(0) {\lrcorner\,} (F^{-1})^*\alpha|_0 = \dot{\gamma}_2(0) {\lrcorner\,} (F^{-1})^*\alpha|_0. This, in turn, implies that v {\lrcorner\,} p_1^*\alpha|_u = v {\lrcorner\,} p_2^* \alpha|_u. Because v \in T_uW is arbitrary, we conclude that p_1^*\alpha|_u = p_2^*\alpha|_u, as required.

Suppose that x is not a fixed point. Then the stabiliser H of x is a proper subgroup of G. Since G is connected, \dim H < \dim G. By the induction hypothesis, every H-basic form on V is the pullback of a differential form on V/H. By Lemma 5.9, every G-basic form on G \times_H V is the pullback of a differential form on (G \times_H V)/G. Because F is an equivariant diffeomorphism, every G-basic form on U is the pullback of a differential form on U/G. So \alpha|_U is the pullback of a differential form on U/G. This implies that p_1^* \alpha|_u = p_2^* \alpha|_u, as required.

y Lemma 3.1 and Proposition 3.2, the pullback is an injection into the space of basic forms. By Remark 2.17 and Proposition 3.4, as a map to its image, the pullback is an isomorphism of differential graded algebras and a diffeological diffeomorphism. By Proposition 5.10, if the identity component of G acts properly, the image is the space of basic forms.