Closed 1-Forms in Topology and Geometric Group Theory

(1) \Longrightarrow (2): Let \overline{X} be the universal abelian cover of X and h:\overline{X}\to{\mathbb R} be obtained from the pullback of \omega to \overline{X}. It is easy to see that we have h(gx)=\xi(g)+h(x) for all x\in \overline{X} and g\in G. We choose N=h^{-1}([0,\infty)), and let N'\subset N be the component such that h is unbounded on N' by [binest, Lemma 5.2]. For every cell \sigma of X pick a lift \bar{\sigma}\subset h^{-1}((-\infty,-1]). If \sigma is a 0-cell, we can find a cellular map H_\sigma:[0,1]\to\overline{X} such that H_\sigma(0)=\bar{\sigma} and H_\sigma(1)\in N'. Here [0,1] has the standard cell structure with two 0-cells and one 1-cell. Using equivariance, this gives an equivariant cellular homotopy H^0:\overline{X}^{(0)}\times [0,1]\to\overline{X}. Note that G acts on \overline{X}\times [0,1] by g(x,t)=(gx,t), and H^0 induces a homotopy on X^{(0)} which gives 1-movability of the 0-skeleton. As the image of H^0_1 is in the 0-skeleton, we can iterate this homotopy to obtain n-movability for any n>0.

Now pick a cell \bar{\sigma}\subset h^{-1}((-\infty,-1]) for every 1-cell \sigma of X and let u,v be the boundary points of \bar{\sigma}. By possibly iterating H^0, we can assume that H^0(u,1),H^0(v,)\in N' and since N' is connected, we can find a cellular map H_{u,v}:[0,1]\to N' connecting these points.

Note that H_{u,v}([0,1]), H^0(\{u,v\}\times [0,1]) and \bar{\sigma} combine to a closed loop in \tilde{X} when suitably oriented. But by Theorem 3 this loop is representable by a loop in N'. In other words, by changing the path H_{u,v} suitably in N', we can assume that this loop bounds. Therefore we can extend H^0 to H_\sigma:(\overline{X}^{(0)}\cup\bar{\sigma})\times [0,1]\to \overline{X} cellularly such that H_\sigma(\bar{\sigma},1)=H_{u,v}([0,1]). Doing this for every 1-cell of X and extending equivariantly gives a cellular and equivariant homotopy H^1:\overline{X}^{(1)}\times [0,1]\to\overline{X} such that H^1_0 is inclusion and h(H^1(x,1))-h(x)\geq 1.

Since H^1 is equivariant we get a homotopy H:X^{(1)}\times [0,1]\to X which shows that X^{(1)} is 1-movable with respect to \omega. By compactness there is a C>0 such that X^{(1)} is 1-movable with control C. Again the image of H_1 is in the 1-skeleton of X, so we can iterate the homotopy. Note that iterating the homotopy does not increase the control, so we get that X^{(1)} is n-movable with respect to \omega with control C for every n>0.

(2) \Longrightarrow (1): Let h:\overline{X}\to{\mathbb R} and N'\subset N=h^{-1}([0,\infty)) be as above. Pick a basepoint x_0\in N' with f(x_0)\geq C+1. Let \gamma:(S^1,1)\to (\tilde{X},x_0) be a loop which can be assumed cellular. By compactness of S^1 there is a K\leq 0 such that \gamma(S^1)\subset f^{-1}([K,\infty)).

By assumption there is a homotopy \bar{H}:\overline{X}^{(1)}\times [0,1]\to\overline{X} with \bar{H}_0 is inclusion and

for all x\in \overline{X}. Let \mu:\tilde{X}\to[0,1] be a map with \mu|f^{-1}([C+1,\infty))\equiv 0 and \mu|f^{-1}((-\infty,C])\equiv 1. Now define A:S^1\times[0,1]\to\tilde{X} by A(x,t)=\bar{H}(\gamma(x),\mu(\gamma(x))\cdot t). Then A(x,0)=\gamma(x), A(x,1)=\bar{H}(\gamma(x),\mu(\gamma(x)))\in N' and A(x_0,t)=x_0 for all t\in[0,1]. Therefore \xi\in\Sigma^1(G;{\mathbb Z}) by Theorem 3.

(1) \Longleftrightarrow (4) as we can lift H to \tilde{H} and \tilde{H} determines H. (2) \Longrightarrow (3) is obvious. (3) \Longrightarrow (4): We define H by induction on the skeleta of \tilde{X}. The homotopy can always be defined on \tilde{X}^{(0)} as \tilde{X} is connected and \xi nonzero. Assume that H:\tilde{X}^{(k-1)}\times [0,1]\to \tilde{X} satisfies the conclusion of (4). Let N=N_0=h_\xi^{-1}([0,\infty)) be a neighborhood of \infty with respect to \xi and N'\subset N as in (3), and choose a \varepsilon>0. Choose a lift \tilde{\sigma}\subset \tilde{X}-N_{-\varepsilon} for every k-cell \sigma of X. By iterating H if necessary, we can assume that H_1(\partial \tilde{\sigma})\subset N'. Given a characteristic map \chi_\sigma:(D^k,S^{k-1})\to (\tilde{X}^{(k)},\tilde{X}^{(k-1)}), we can compose \chi_\sigma|S^{k-1} with H to get an element of \pi_k(\tilde{X},N') which restricts to the trivial element of \pi_k(\tilde{X},N). This gives a homotopy \chi:D^k\times[0,1]\to\tilde{X} with \chi_0=\chi_\sigma and \chi_1(D^k)\subset N. We can use this homotopy to extend H equivariantly to the k-skeleton such that (4) is satisfied.

To see that (4) \Longrightarrow (2) observe that H can be used to homotop every map \varphi:(D^k,S^{k-1})\to (\tilde{X},N_i) to a map with image in any neighborhood of \infty. As the base point should not be moved during the homotopy, we have to modify H on a ‘buffer zone’ N_{i-\lambda}-N_i so that points mapped to N_i will not be changed. Nevertheless we can find N_{i-\lambda} such that every \varphi can be homotoped to a map (D^k,S^{k-1})\to (N_{i-\lambda}) which gives (1).

Let h:\tilde{X}\to {\mathbb R}^r by the composition of the covering map \bar{p}:\tilde{X}\to\overline{X} with the Abel-Jaobi map A:\overline{X}\to {\mathbb R}^r. Then, given \xi:\pi_1(X)\to{\mathbb R} we get a height function by h_\xi=l_\xi\circ h where l_\xi:\pi_1(X)/[\pi_1(X),\pi_1(X)]\otimes {\mathbb R}\to {\mathbb R} is defined by l_\xi([g]\otimes t)=\xi(g)\cdot t. Now let \tilde{H} be a homotopy as in Proposition 2 (4) for a \xi. Define \tilde{K}:S(X)\times \tilde{X}^{(k)}\to{\mathbb R} by

By the choice of H we get \tilde{K}(\xi,\tilde{x})\geq \varepsilon for all \tilde{x}\in\tilde{X} and some \varepsilon>0. It also induces a map K:S(X)\times X^{(k)} \to {\mathbb R}. From the compactness of X^{(k)}, we get

for all \tilde{x}\in\tilde{X}^{(k)}and all \xi' in a neighborhood of \xi in S(X). Therefore Proposition 2 (4) is satisfied for all such \xi'.

It is easy to see that Condition (3) of Proposition 2 implies Condition (3) of Proposition 4 so that \Sigma^k(X)\subset \Sigma^k(X,{\mathbb Z}) for k\geq 1. Also, for k=1 the chain homotopy of Proposition 4 (3) can easily be used to realize a homotopy as in Proposition 2 (3).

(1) \Longrightarrow (2): Let A:C_\ast(\tilde{X})\to C_\ast(\tilde{X}) be the chain map chain homotopic to the identity, given by Proposition 4 (4). Then {\rm id}-A:C_i(X;\widehat{{\mathbb Z} G}_{-\xi})\to C_i(X;\widehat{{\mathbb Z} G}_{-\xi}) is an isomorphism with inverse {\rm id}+A+A^2+\ldots, which converges over \widehat{{\mathbb Z} G}_{-\xi} by the valuation property in Proposition 4(4). So the map on homology is both an isomorphism and zero, which means that the homology vanishes.

(2) \Longrightarrow (1): As C_\ast(X;\widehat{{\mathbb Z} G}_{-\xi}) is free and bounded below, the vanishing of its homology groups up to degree k guarantees the existence of a chain homotopy \delta:C_\ast(X;\widehat{{\mathbb Z} G}_{-\xi})\to C_{\ast+1}(X;\widehat{{\mathbb Z} G}_{-\xi}) with \partial_{i+1}\delta_i+\delta_{i-1}\partial_i={\rm id} for i\leq k. “Cutting off” gives a chain homotopy \bar{\delta}:C_\ast(\tilde{X})\to C_\ast(\tilde{X}) with \partial_{i+1}\bar{\delta}_i+\bar{\delta}_{i-1}\partial_i={\rm id}-A. Then A is a chain map homotopic to the identity, and by approximating \delta sufficiently well with \bar{\delta}, condition (4) of Proposition 4 is satisfied.

Let U be a small open neighborhood of X^{(k)} in X which deformation retracts to X^{(k)}. Let H:X^{(k)}\times [0,1]\to X be a homotopy starting with inclusion which lifts to a homotopy as in Proposition 2 (3). Now define h:X^{(k)}\times [0,\infty)\to X by h(x,t)=H(H^m(x,1),t-m), where m is an integer with t\in[m,m+1]. Notice that H is cellular so H^m(x,1)\in X^{(k)}. It is easy to see that h combined with the deformation retraction of U to X^{(k)} gives a homotopy as required in Definition 12. It is well known that {\rm {cat }}_X(X-X^{(k)})\leq \dim(X)-k so the result follows.

Notice that

if every 1-sphere in N_i bounds in N_{i-\lambda}, in particular the images become abelian. As \xi\in\Sigma^2(\pi_1(X)), we can define a homomorphism \pi_2(\tilde{X},N_i)\to H_2(\tilde{X},N_{i-\lambda}) as in a typical proof of the classical Hurewicz Theorem (after possibly increasing \lambda), see, for example, Spanier [spanie]. The details will be left to the reader.

This is similar to the proof of Lemma 8, but easier, as we can define the homomorphism directly.

We give a proof for the homotopy invariant \Sigma^k(Y). Choose liftings \tilde{f}:\tilde{X}\to\tilde{Y} and \tilde{g}:\tilde{Y}\to \tilde{X} with \tilde{f}\tilde{g}\simeq {\rm id}_{\tilde{Y}} equivariantly. Let \xi\in S(Y) satisfy f^\ast(\xi)\in \Sigma^k(X). Let h_\xi:\tilde{Y}\to {\mathbb R} be a height function. Then h_\xi\circ \tilde{f}:\tilde{X}\to{\mathbb R} is a height function for f^\ast\xi.

By cocompactness, there exists a C\geq 0 such that

for every \tilde{y}\in \tilde{Y}.

By assumption, Proposition 2 and Remark 1, there is an equivariant homotopy \tilde{H}:\tilde{X}^{(k)}\times [0,1]\to \tilde{X} starting with inclusion, such that

for all \tilde{x}\in\tilde{X}. Therefore

Now \tilde{f}H(\tilde{g},\cdot) can be combined with the homotopy \tilde{f}\tilde{g}\simeq {\rm id}_{\tilde{Y}} to show that \xi\in\Sigma^k(Y).

Add cells of dimension \geq m+1 to X to get a (possibly infinite) CW-complex X' containing X such that f extends to a homotopy equivalence f':X'\to Y. Let g:Y\to X' be a cellular homotopy inverse. Then g(Y^{(m)})\subset X. If f^\ast(\xi)\in\Sigma^k(X), the homotopy H:\tilde{X}^{(k)}\times [0,1]\to \tilde{X} can now be used as in the proof of Proposition 10 to show that \xi\in\Sigma^k(Y) for k\leq m.

Let h:\bar{X}\to{\mathbb R} be induced by a height function h_\xi:\tilde{X}\to{\mathbb R} and \zeta:\bar{X}\to\bar{X} the generating covering transformation with h\zeta(x)>h(x) for all x\in\bar{X}. Let a:Y\to \bar{X} and b:\bar{X}\to Y be mutually inverse homotopy equivalences. The there is a homotopy equivalence g:M_{b\zeta a}\to X given by

where the last homotopy equivalence is given by [\bar{x},t]\to q(\bar{x}) for \bar{x}\in\bar{X}.

We can assume that b\zeta a:Y\to Y sends the k-skeleton to the k-skeleton. Let \varphi:Y^{(k)}\to Y^{(k)} be the restriction of b\zeta a to Y^{(k)}. The induced map M_\varphi\to M_{b\zeta a} is k-connected and so there is a k-connected map f:M_\varphi\to X. By Example 2 -f^\ast(\xi)\in \Sigma^k(M_\varphi). It follows from Proposition 12 that -\xi\in\Sigma^k(X). To get \xi\in \Sigma^k(X) as well, replace \zeta by \zeta^{-1}.

If C is {\mathbb Z} N-chain homotopy equivalent to a finitely generated projective {\mathbb Z} N-complex, it is, by definition, of finite n-type over {\mathbb Z} N, hence S(G,N)\subset \Sigma^n(C) by Theorem 8.

If S(G,N)\subset \Sigma^n(C), then by Lemma ? there is a chain complex D', finitely generated free over {\mathbb Z} N with D_i'=0 for i>n which dominates C. By [ranims, Prop.3.1], the required chain complex D exists.

If \mathcal{M}_\xi is (k-1)-connected, then by Proposition 20 we get that the inverse system \{\pi_l(\tilde{X},N)\} is pro-trivial for l\leq k which gives \xi\in \Sigma^k(X) by Proposition 2.

To get the other direction, we have to worry about {\lim\limits_\leftarrow}^1\,\pi_{k+1}(\tilde{X},N_i). But by the next lemma \{\pi_{k+1}(\tilde{X},N)\} is semi-stable, so the {\lim\limits_\leftarrow}^1-term vanishes. It follows from Proposition 20 that \mathcal{M}_\xi is (k-1)-connected.

Use the homotopy from Proposition 2(4) to push any k-sphere in N arbitrarily far away, this gives semi-stability as the pushing may be done in N' slightly bigger than N (depends on H only).

The proof is similar to the proof of [brown2, Thm.2.2], but one has to use Proposition 23. We omit the details.