Adiabatic limits and spectral sequences for Riemannian foliations

The equalities involving d_{1,0} in property (i) follow from (?), (?) since

The other equalities in property (i) are obtained by taking adjoints, and property (ii) is a direct consequence of property (i).

Set D_\perp=d_{1,0}+\delta_{-1,0}. Then, by Remark 3.7 and the proof of Lemma 7.2 in [AlvKordy1], the operators

on \Omega define bounded operators on \boldsymbol{\Omega}. This easily yields property (i). Now properties (ii) and (iii) follows from property (i) since d_{2,-1} and \delta_{-2,1} are of order zero, and d_{0,1} and \delta_{0,-1} vanish on {\text{H}}_1 and preserve each \widetilde{\text{H}}_1^{\cdot,v}.

For r=u+v we have

Then

by (?), (?) and (?), and the result follows.

Take any \alpha\in D^{u,v}. For r=u+v, there exists some \beta\in\boldsymbol{\Omega}_{u+1}^r such that \alpha+\beta\in\operatorname{dom}{\mathbf d}. So \pi_{u,v}{\mathbf d}(\alpha+\beta) is defined in \boldsymbol{\Omega}^{u,v}. But \pi_{u,v}{\mathbf d}(\alpha+\beta)={\mathbf d}_{0,1}\alpha because \alpha+\beta\in\boldsymbol{\Omega}_u^r.

For r=u+v, we have

Since the orthogonal projection

preserves smoothness on M, the result follows by (?) and (?).

Take any \alpha\in D^{u,v}\cap\operatorname{cl}_0\left(d_{0,1}\Omega^{u,v-1}\right), and fix some \beta\in \boldsymbol{\Omega}_{u+1}^r with \alpha+\beta\in\operatorname{dom}{\mathbf d}, where r=u+v. We know that \pi_{u+1,v}{\mathbf d}(\alpha+\beta)\in D^{u+1,v} by Lemma ?. On the other hand, if \bar{\mathbf d}_{0,1} and \bar{\mathbf d}_{1,0} denote the extensions of d_{0,1} and d_{1,0} to continuous maps \boldsymbol{\Omega}\to W^{-1}\Omega, we have

where \beta_1=\pi_{u+1,v-1}\beta\in \boldsymbol{\Omega}^{u+1,v-1}, and

Hence

by Theorem 2.1. Therefore the result follows by Lemma ? and Corollary ?, ?.

The inclusion “\supset” of the first equality is obvious, and the inclusion “\supset” of the second equality follows from Lemma ?-(iii).

To prove the inclusion “\subset” of the first equality, by (?) it is enough to prove that \Pi\alpha\in D^{u,v} for all \alpha\in D^{u,v}\cap\ker{\mathbf d}_{0,1}. This obviously holds if we prove \Pi\alpha\in\operatorname{dom}{\mathbf d}_1 for every such an \alpha since the inclusion “\supset” of the second equality is already proved. This also proves the inclusion “\subset” of the second equality by taking \alpha\in L^2\text{H}_1.

Thus take any \alpha\in D^{u,v}\cap\ker{\mathbf d}_{0,1}. Then there is some \beta\in \boldsymbol{\Omega}_{u+1}^r such that \alpha+\beta\in\operatorname{dom}{\mathbf d}, where r=u+v. Write \beta=\beta_1+\beta_2 with \beta_1\in \boldsymbol{\Omega}^{u+1,v-1} and \beta_2\in \boldsymbol{\Omega}_{u+2}^r. Thus, since \alpha\in\ker{\mathbf d}_{0,1}, we get

Here we consider

\Pi and \pi_{u+1,v} as bounded operators on W^{-1}\Omega. But

because Q\alpha=0, and

by Lemma ?-(i). Therefore \Pi\bar{\mathbf d}_{1,0}\Pi\alpha\in\boldsymbol{\Omega}, yielding \Pi\alpha\in\operatorname{dom}{\mathbf d}_1 as desired.

This follows from Corollary ?, ? and Lemma ?.

This follows from Corollary ? and Theorem 2.2-(iii),(iv).

This follows from Theorem 2.2-(iv) and Corollary ?.

The result follows from (?) and (Proof) with easy arguments (see Lemma 2.5 in [Sergiescu87] and Lemma 7.4 in [AlvKordy1]).

This easily follows from (?) and (?).

The inclusion “\supset” of the first equality is obvious, and the inclusion “\supset” of the second equality follows from Lemma ?-(iii).

To prove the inclusion “\subset” of the first equality, by (?) it is enough to prove that \widetilde\Pi\alpha\in\widetilde{D}^{u,v} for all \alpha\in \widetilde{D}^{u,v}\cap\operatorname{cl}_0({\text{B}}_v). This obviously holds if we prove \widetilde\Pi\alpha\in\operatorname{dom}\tilde{\mathbf d}_1 for every such an \alpha since the inclusion “\supset” of the second equality is already proved. This also proves the inclusion “\subset” of the second equality by taking \alpha\in L^2\widetilde\text{H}_1.

Thus take any \alpha\in \widetilde{D}^{u,v}\cap\operatorname{cl}_0({\text{B}}_v). Then there is some \beta\in \boldsymbol{\Omega}_{u+2}^r such that \alpha+\beta\in\operatorname{dom}{\mathbf d}, where r=u+v. Write \alpha=\alpha_1+\alpha_2 with \alpha_1\in \boldsymbol{\Omega}^{u,v} and \alpha_2\in \boldsymbol{\Omega}^{u+1,v-1}. So, since \alpha\in\operatorname{cl}_0({\text{B}}_v) and \bar{\mathbf d}\beta\in\ker\widetilde\Pi_{\cdot,v}, where \widetilde\Pi_{\cdot,v} is considered as a projection in W^{-1}\Omega, we get

But

because \alpha\in \widetilde{D}^{u,v}\cap\operatorname{cl}_0({\text{B}}_v), and

by Lemma ?-(ii). Therefore \widetilde\Pi_{\cdot,v}\bar{\mathbf d}\widetilde\Pi_{\cdot,v}\alpha\in\boldsymbol{\Omega}, yielding \widetilde\Pi_{\cdot,v}\alpha\in\operatorname{dom}\tilde{\mathbf d}_1 as desired.

This follows from Lemmas ? and ?.

This follows from (?), (?), Lemma ?, Corollary ? and Theorem 2.2-(v).

The canonical map H({\mathbf e}_1)\to H(\hat{\mathbf e}_1) is a linear isomorphism by Corollary ? and the exactness of (?). Moreover it is obviously continuous. Then it is also an homeomorphism because H(\hat{\mathbf e}_1) is a Hausdorff topological vector space of finite dimension.

By the commutativity of the diagram

where all maps are canonical, the result follows directly from Theorem 2.2-(vii), Corollaries ? and ?.

Consider the compositions

where the first map of each composition is canonical, and the second one is canonically induced by the projections \pi_{u,v}. The first composition is an isomorphism of topological vector spaces by Theorem 2.2-(vii), and we know that the second composition is a continuous linear isomorphism (Section ?). Then the second composition is also an homeomorphism because H({\mathbf e}_1) is Hausdorff of finite dimension by Corollary ?. So the result follows from Corollary ? and the commutativity of the diagram

where the horizontal arrows denote the above compositions, and the vertical arrows denote canonical maps.

This is a direct consequence of Lemma ?.

This follows from Corollary ? since each d_\ell induces isomorphisms

>From Lemma ? we get a (non-canonical) isomorphism

Then let L_k^{u,v} be the subspace of \text{A}^r/(\text{A}^r\cap\ker d) that corresponds to

by (Proof). Then property (i) easily follows from Lemma ?, and property (ii) is obvious; in fact, L_k^r corresponds to

by (Proof).

We have (V+W)\cap\overline{W}=W since W is closed in V+W, yielding V\cap\overline{W}=0 because V\cap W=0. So V+\overline{W}=V\oplus\overline{W} as topological vector spaces because all spaces involved are closed subspaces of L (see for instance [Schaefer, Corollary 3 of Theorem 2.1, Chapter III, page 78]). Now the result follows easily.

The space \boldsymbol{\Omega}^r\cap\ker{\mathbf d} is closed in \boldsymbol{\Omega} since {\mathbf d} is a closed operator, and thus so is its subspace {\mathbf Z}_\infty^{u,v}=\boldsymbol{\Omega}_u\cap(\boldsymbol{\Omega}^r\cap\ker{\mathbf d}). Hence \boldsymbol{\Omega}^r\cap\ker{\mathbf d}=V\oplus{\mathbf Z}_\infty^{u,v} as Hilbert spaces, where V is the orthogonal complement of {\mathbf Z}_\infty^{u,v} in \boldsymbol{\Omega}^r\cap\ker{\mathbf d}; in particular V is closed in \boldsymbol{\Omega} too. Obviously,

On the other hand we clearly have

and thus {\mathbf Z}_\infty^{u,v}+\boldsymbol{\Omega}_{u+1}^r and {\mathbf Z}_k^{u,v}+\boldsymbol{\Omega}_{u+1}^r are respectively closed in (\boldsymbol{\Omega}^r\cap\ker{\mathbf d})+\boldsymbol{\Omega}_{u+1}^r and {\mathbf Z}_k^{u,v}+(\boldsymbol{\Omega}^r\cap\ker{\mathbf d})+\boldsymbol{\Omega}_{u+1}^r. Therefore Lemma ? yields

as topological vector spaces, and the result follows.

This follows from Lemma ? since we clearly have

as topological vector spaces.

This follows directly from Corollary ? and (?).

Let u+v=r. The restriction \rho_u^r:{\mathbf Z}_k^{u,v}+(\boldsymbol{\Omega}^r\cap\ker{\mathbf d})\to X_u^r induces a homomorphism

We clearly have

So \bar\rho_u^r induces a continuous linear isomorphism

Observe that \operatorname{im}\bar\rho_u^r is a Hausdorff topological vector space by Corollary ?, and thus \left\|\cdot\right\| and \left\|\cdot\right\|_h induce norms on \operatorname{im}\bar\rho_u^r that will be also denoted by \left\|\cdot\right\| and \left\|\cdot\right\|_h, respectively. By (Proof) and Corollary ?, the homomorphism \bar\rho_u^r restricts to an injection \bar\rho_u^r:L_k^{u,v}\to\operatorname{im}\bar\rho_u^r. Since \rho_u^r is an orthogonal projection for any metric g_h, we easily get

Here, we use the norm on \operatorname{im}\bar\rho_u^r in the left hand side of (Proof), and the norm on \boldsymbol{\Omega}/\ker{\mathbf d} in its right hand side. Observe that, by (?),

yielding

Moreover, since K is of finite dimension, \operatorname{im}\bar\rho_u^r is Hausdorff, and the restriction \bar\rho_u^r:L_k^{u,v}\to\operatorname{im}\bar\rho_u^r is injective, we get the existence of some C'_K>0 so that

So

for all \zeta\in K and 0<h\leq 1 as desired.

Since K is of finite dimension, there is some constant C''_K, depending on K, so that

Because L_k^r=\bigoplus_{u+v=r}L_k^{u,v}, any finite dimensional subspace K\subset L_k^r is contained in the sum of finite dimensional subspaces K^{u,v}\subset L_k^{u,v}, u+v=r. Therefore we can assume K is contained in some L_k^{u,v} with u+v=r. Then, for \zeta\in K and 0<h\leq 1, we have

and the result follows with C_K=C'_KC''_K.

The result is proved by induction on k. First, the case k=1 is trivial.

Second, the kernel and the image of d_1 in e_1^{u,v} respectively are z_2^{u,v}/b_0^{u,v} and b_1^{u,v}/b_0^{u,v}, whose canonical projections in \hat e_1^{u,v}=z_1^{u,v}/\overline{b_0^{u,v}} are

yielding the canonical isomorphism (?), (?), (?) for k=2. Since the isomorphism e_2^{u,v}\cong\hat e_{1,2}^{u,v} is canonically defined, it corresponds to the canonical map (?), (?), (?) for k=2.

Now assume the result holds for k=\ell\geq2 and we prove it for k=\ell+1. The kernel and the image of d_\ell in e_\ell^{u,v} respectively are z_{\ell+1}^{u,v}/b_{\ell-1}^{u,v} and b_\ell^{u,v}/b_{\ell-1}^{u,v}, whose images by the canonical isomorphism (?), (?), (?) for k=\ell are

These spaces respectively correspond to the kernel and the image of d_\ell in \hat e_{1,\ell}^{u,v} by (?), (?), (?), yielding the canonical isomorphism (?), (?), (?) for k=\ell+1. Again, because the isomorphism e_{\ell+1}^{u,v}\cong\hat e_{1,\ell+1}^{u,v} is canonically defined, it is given by the canonical map (?), (?), (?) for k=\ell+1.

Observe that the first part of the statement means that we have an orthogonal decomposition

Again the result follows by induction on k. We have an orthogonal decomposition

by Theorem 2.1. Thus the isomorphism \hat e_{1,1}=\hat e_1^{u,v}\stackrel{\cong}{\to}\text{H}_1^{u,v} is induced by the orthogonal projection \Pi_1 onto \text{H}_1. On the other hand, the kernel and image of d_1 in \text{H}_1^{u,v} respectively correspond by this isomorphism to the kernel and image of d_1 on \hat e_1^{u,v}, which are respectively given by (Proof). So the kernel and image of d_1 in \text{H}_1^{u,v} are the orthogonal projections \Pi_1\left(z_2^{u,v}+\overline{b_0^{u,v}}\right) and \Pi_1\left(b_1^{u,v}+\overline{b_0^{u,v}}\right), respectively. Hence, by definition, \text{H}_2^{u,v} is the orthogonal complement of \Pi_1\left(b_1^{u,v}+\overline{b_0^{u,v}}\right) in \Pi_1\left(z_2^{u,v}+\overline{b_0^{u,v}}\right), which is equal to the orthogonal complement of b_1^{u,v}+\overline{b_0^{u,v}} in z_2^{u,v}+\overline{b_0^{u,v}} by (?) since

Thus the result follows for k=2.

Now suppose the statement holds for k=\ell\geq2. Then, via the isomorphism \hat e_{1,\ell}^{u,v}\stackrel{\cong}{\to}\text{H}_\ell^{u,v} induced by \Pi_\ell, the kernel and image of d_\ell in \text{H}_\ell^{u,v} respectively correspond to the kernel and image of d_\ell in \hat e_{1,\ell}^{u,v}, which are given in (?). So the kernel and image of d_\ell in \text{H}_\ell^{u,v} are the orthogonal projections \Pi_\ell\left(z_{\ell+1}^{u,v}+\overline{b_0^{u,v}}\right) and \Pi_\ell\left(b_\ell^{u,v}+\overline{b_0^{u,v}}\right), respectively. Hence, by definition, \text{H}_{\ell+1}^{u,v} is the orthogonal complement of \Pi_\ell\left(b_\ell^{u,v}+\overline{b_0^{u,v}}\right) in \Pi_\ell\left(z_{\ell+1}^{u,v}+\overline{b_0^{u,v}}\right), which is equal to the orthogonal complement of b_\ell^{u,v}+\overline{b_0^{u,v}} in z_{\ell+1}^{u,v}+\overline{b_0^{u,v}} by (Proof) for k=\ell since

Thus the result follows for k=\ell+1.

Write \beta=\beta'+\beta'' with \beta'\in P(\Omega^{u-1,v}) and \beta''\in Q(\Omega^{u,v-1}). Then

by (?), (?), Lemma ?, ? and because \Pi_2d_{0,1}=\Pi_2d_1=\Pi_2P=0.

Since the image of \tilde d_1 is closed and equal to its kernel, the hypothesis implies the existence of a sequence \beta_i\in\widetilde\text{H}_1^{u-1,v} such that \alpha_i+\tilde d_1\beta_i\to0 strongly in \boldsymbol{\Omega}. On the other hand we have

on \widetilde\text{H}_1^{u,v}, and thus the operator \Pi_2\pi_{u+2,v-1}d:\widetilde\text{H}_1^{u,v}\to\text{H}_2^{u+2,v-1} is bounded because d_{2,-1} and \Pi d_{1,0}Q are bounded operators in \boldsymbol{\Omega} by Lemma ?. Therefore

strongly in \boldsymbol{\Omega}. Then the result follows directly from Lemma ?, ?.

Consider the following subspaces of \Omega^{u+v}:

First, observe that each \widetilde\text{H}_\ell^{u,v} is the orthogonal complement of \text{A}_{\ell-1}^{u+\ell,v-\ell} in \overline{\text{B}^{u,v}}, and thus \widetilde\text{H}_\ell^{\cdot,v} is the orthogonal complement of \text{A}_{\ell-1}^{\cdot,v-\ell}=\bigoplus_a\text{A}_\ell^{a,v-\ell} in \overline{\text{B}^{\cdot,v}}=\bigoplus_a\overline{\text{B}^{a,v}}. So the inclusion \widetilde\text{H}_\ell^{\cdot,v}\hookrightarrow\overline{\text{B}^{\cdot,v}} induces an isomorphism of topological vector spaces

whose inverse is induced by the orthogonal projection \widetilde\Pi_{\cdot,v}:\overline{\text{B}^{\cdot,v}}\to\widetilde\text{H}_\ell^{\cdot,v}. Second, observe that both \overline{\text{B}^{u,v}} and \text{A}_{\ell-1}^{\cdot,v-\ell} are subcomplexes of (\Omega,d). Moreover \tilde d_\ell in \widetilde\text{H}_\ell^{\cdot,v} clearly corresponds to the differential map in the quotient complex \overline{\text{B}^{\cdot,v}}/\text{A}_{\ell-1}^{\cdot,v-\ell} via (?), (?), (?), (?), and the result follows.

By the expression

it is enough to consider the following three cases. First, assume \beta\in\bigoplus_{a<\ell}\Omega^{u-1+a,v-a} and let \beta'=\pi_{u+\ell-2,v-\ell+1}\beta. We clearly have

yielding

by Lemma ?, ?, and because \Pi_{\ell+1}d_1=0 and \Pi_{\ell+1}P_0=0.

Second, suppose \beta\in\widetilde\text{H}_1^{u+\ell-2,v-\ell+1} and write \beta=\beta'+\beta'' with

We clearly have

yielding

by Lemma ?, ?.

Third, assume \beta\in(Q_1+\cdots+Q_{\ell-1})(\Omega^{u+\ell-1,v-\ell}), which is contained in \text{H}_1^{u+\ell-1,v-\ell}. Then the result follows because \widetilde\Pi_{\ell;\cdot,v}d=\widetilde\Pi_{\ell;\cdot,v}d_1 on \text{H}_1^{\cdot,v-\ell}, and

Write \alpha=\alpha'+\alpha''+\alpha''' with \alpha'\in P_0(\Omega^{u,v}), \alpha''\in\Omega^{u+1,v-1} and

Observe that \alpha'+Q_0\alpha''\in\widetilde\text{H}_1^{\cdot,v}. Since \tilde d_\ell\alpha=0 and \tilde d_1=\widetilde\Pi_{1;\cdot,v}\tilde d_\ell on \widetilde\text{H}_1^{\cdot,v}, we have \tilde d_1(\alpha'+Q_0\alpha'')=0. Thus there is some \beta\in\widetilde\text{H}_1^{u-1,v} with \tilde d_1\beta=\alpha'+Q_0\alpha'' because H(\widetilde\text{H}_1^{\cdot,v},\tilde d_1)=0. Then \alpha-\tilde d_\ell\beta\in\widetilde\text{H}_\ell^{u,v} satisfies

and moreover

by Lemma ?, ?. Therefore we can assume \alpha'+Q_0\alpha''=0, and thus \alpha'=Q_0\alpha''=0. With this assumption, it follows that \alpha''=(\Pi_1+P_0)\alpha'' and

by Lemma ?, ?, yielding Q_1\alpha''=0.

Take a sequence

such that d_{0,1}\phi_i is C^\infty convergent to P_0\alpha''. Then the sequence \alpha-\tilde d_\ell\phi_i\in\widetilde\text{H}_\ell^{u,v} satisfies

by Lemma ?, ?. So

and thus we can also assume P_0\alpha''=0.

For each k=1,\dots,\ell-1 there is some \sigma^k\in Q_k(\Omega^{u-k+1,v+k-2}) with d_k\sigma^k=P_k\alpha''. As above, from the existence of such a \sigma^1 we can assume P_1\alpha''=0 by Lemma ?, ? since d_1=\pi_{u,v-1}\Pi_1\tilde d_\ell on \text{H}_1^{u,v-1}. If \ell>2, by Proposition ? for k=2,\ldots,\ell-1 there is a sequence \tau^k_i\in\Omega^{u+v-1}_{u-m+2} such that

with respect to the C^\infty topology in \Omega. We can thus suppose P_k\alpha''=0 for such a k because

by Lemma ?, ?. Therefore

where the last term does not show up if \ell=2.

Now the condition \tilde d_\ell\alpha=0 can be written as

Observe that (?) summarizes the conditions of the first part of Proposition ? for k=2,\dots,\ell, with \omega=\alpha'', the constant sequence \alpha_i=\alpha''', and \gamma=0 if 2\leq k<\ell. Since \alpha''\in\text{H}_2^{u+1,v-1} by (?), we get inductively on k=2,\dots,\ell-1 that \alpha''\in\text{H}_k^{u+1,v-1} and d_k\alpha''=0 by (?), (?) and Proposition ?. Hence \alpha''\in\text{H}_\ell^{u+1,v-1} by (?), and thus

by (?) and Proposition ? for k=\ell, and the result follows.

By Theorem 2.2 in [Chernoff], D=d+\delta is essentially self-adjoint in \boldsymbol{\Omega}. Then, by using e.g. Lemma XII.1.6–(c) in [DunfordSchwartz], so is \widetilde\Pi_{\ell;\cdot,v} D\widetilde\Pi_{\ell;\cdot,v} because \widetilde\Pi_{\ell;\cdot,v} is a bounded self-adjoint operator on \boldsymbol{\Omega}. But \widetilde\Pi_{\ell;\cdot,v} D\widetilde\Pi_{\ell;\cdot,v} is equal to \widetilde D_\ell in L^2\widetilde\text{H}_\ell^{\cdot,v} and vanishes in its orthogonal complement. Hence \widetilde D_\ell is essentially self-adjoint.

We have

But

on \widetilde\text{H}_\ell^{\cdot,v}. Then the result follows by Lemma ?-(i).

Corollary ? implies that each inclusion W^{r+1}\widetilde\text{H}_\ell^{\cdot,v}\hookrightarrow W^r\widetilde\text{H}_\ell^{\cdot,v} is a compact operator, and \bigcap_rW^r\widetilde\text{H}_\ell^{\cdot,v}=\widetilde\text{H}_\ell^{\cdot,v}. Then the result follows by Proposition 2.44 in [AlvTond1] and Lemma ?.

We can assume that M is oriented by using the two fold covering of orientations with standard arguments. Then it is easy to check that the Hodge star operator, \star:\Omega\to\Omega, satisfies \star\text{H}_k=\text{H}_k, and \star d_k=(-1)^{r+1}\delta_k\star on \text{H}_k^r for each integer r. Then the result follows from Proposition ?.

Consider the operators B,F given by Lemma 5.16 and (?). Since B,F are of order zero, there is some C'>0 such that B^\ast B,F^2\leq C'. Because D_0 is symmetric, we get

for all \omega\in\Omega, yielding

Similarly we get

Therefore, from (?) and Lemma 5.16 we get

for some C>0 and all h small enough.