ScalingStacks

6.5 Projectors and homotopies in Morse theory [03SC]

Original official author HTML, exact retained edition. Historical TeX conversion verdicts remain unchanged. Cited-edition alignment and mathematical self-containment are not assessed.

Complete original source context · Original author HTML

6.5 Projectors and homotopies in Morse theory

We would like to apply formulas for the A∞A_{\infty}-structure on a subcomplex to the proof of the Theorem 2. In order to do that we need to identify the Morse complex with a direct summand of the de Rham complex. Our approach is based on the ideas of Harvey and Lawson (see [HL]).

Let YY be a compact oriented smooth manifold, d​i​m​Y=ndim\,Y=n. The space of currents D′​(Y)D^{\prime}(Y) we will identify with the space of distribution-valued differential forms. Continuous linear operators Ω∗​(Y)→D′​(Y)\Omega^{\ast}(Y)\to D^{\prime}(Y) are given by their Schwartz kernels, which are elements of D′​(Y×Y)D^{\prime}(Y\times Y). Smoothening operators D′​(Y)→Ω∗​(Y)D^{\prime}(Y)\to\Omega^{\ast}(Y) have kernels in Ω∗​(Y×Y)⊂D′​(Y×Y)\Omega^{\ast}(Y\times Y)\subset D^{\prime}(Y\times Y).

With any oriented submanifold Z⊂YZ\subset Y , dimZ=k\dim\,Z=k of finite volume we associate a canonical current [Z][Z] of degree n−kn-k (namely, we can integrate smooth kk-forms over ZZ).

Let gYg_{Y} be a Riemannian metric on YY, and ff be a Morse-Smale function. The gradient flow e​x​p​(t​g​r​a​d​(f)),t≥0exp(t\,grad(f)),t\geq 0 gives rise to a 11-parameter semigroup acting on Ω∗​(Y)\Omega^{\ast}(Y): ψt​(α)=e​x​p​(t​g​r​a​d​(f))∗​(α)\psi^{t}(\alpha)=exp(t\,grad(f))_{\ast}(\alpha). Schwartz kernel of ψt\psi^{t} is [Gt][G_{t}] where manifold Gt⊂Y×YG_{t}\subset Y\times Y is given by Gt:=g​r​a​p​h​(e​x​p​(t​g​r​a​d​(f)))G_{t}:=graph(exp(t\,grad(f))). We also have the identity

i​d−ψt=d​Ht+Ht​d,id-\psi^{t}=dH^{t}+H^{t}d,

where Ht:Ω∗​(Y)→Ω∗​(Y)⊂D′​(Y)H^{t}:\Omega^{\ast}(Y)\to\Omega^{\ast}(Y)\subset D^{\prime}(Y) is a linear operator of degree −1-1 defined by the distributional kernel [Zt][Z_{t}], Zt:=∪0≤t′≤tgraph(exp(t′grad(f)))Z_{t}:=\cup_{0\leq t^{\prime}\leq t}graph(exp(t^{\prime}\,grad(f))).

It is checked in [HL] that this picture has a limit (in certain sense) as t→+∞t\to+\infty. Namely, there exist limits of currents [Gt][G_{t}] and [Zt][Z_{t}]:

[G∞]=l​i​mt→+∞​[Gt]=∑x∈C​r​(f)[Sx]×[Ux][G_{\infty}]=lim_{t\to+\infty}[G_{t}]=\sum_{x\in Cr(f)}[S_{x}]\times[U_{x}]
[Z∞]=limt→+∞[Zt]=[∪0≤t<+∞Gt][Z_{\infty}]=lim_{t\to+\infty}[Z_{t}]=[\cup_{0\leq t<+\infty}G_{t}]

Linear operators ψ∞\psi^{\infty} (of degree zero) and H∞H^{\infty} (of degree −1-1), corresponding to these kernels, map Ω∗​(Y)\Omega^{\ast}(Y) to D′​(Y)D^{\prime}(Y) and satisfy the identity

i−ψ∞=d​H∞+H∞​d,i-\psi^{\infty}=dH^{\infty}+H^{\infty}d,

where i:Ω∗​(Y)→D′​(Y)i:\Omega^{\ast}(Y)\to D^{\prime}(Y) is the natural inclusion. According to the de Rham theorem this inclusion is a quasi-isomorphism of complexes, therefore ψ∞\psi^{\infty} is. Morally, Π∞:=ψ∞\Pi_{\infty}:=\psi^{\infty} should be thought of as a projector. The image Π∞​(Ω∗​(Y))⊂D′​(Y)\Pi_{\infty}(\Omega^{\ast}(Y))\subset D^{\prime}(Y) coincides with ⊕x∈C​r​(f)𝐑⋅[Ux]\oplus_{x\in Cr(f)}{{\bf R}}\cdot[U_{x}]. We have

Π∞​(α)=∑x∈C​r​(f)(∫Sxα)⋅[Ux]=∑x∈C​r​(f)∫Y(α∧[Sx])⋅[Ux].\Pi_{\infty}(\alpha)=\sum_{x\in Cr(f)}(\int_{S_{x}}\alpha)\cdot[U_{x}]=\sum_{x\in Cr(f)}\int_{Y}(\alpha\wedge[S_{x}])\cdot[U_{x}].

Moreover, the operator Π∞\Pi_{\infty} commutes with the differentials. Hence the complex Π∞​(Ω∗​(Y))\Pi_{\infty}(\Omega^{\ast}(Y)) is a finite-dimensional subcomplex of D′​(Y)D^{\prime}(Y) isomorphic to the Morse complex M∗​(Y,f)M^{\ast}(Y,f). In fact it is quasi-isomorphic to both complexes Ω∗​(Y)\Omega^{\ast}(Y) and D′​(Y)D^{\prime}(Y). In this way Harvey and Lawson prove that the de Rham cohomology is isomorphic to the cohomology of Morse complex.

In order to construct actual projectors and homotopies we will proceed as follows. Let ρδ,δ→0\rho_{\delta},\delta\to 0 be a family of smooth closed differential nn-forms on Y×YY\times Y such that s​u​p​p​(ρδ)supp(\rho_{\delta}) belongs to the open δ\delta-neighborhood NδN_{\delta} of the diagonal d​i​a​g⊂Y×Ydiag\subset Y\times Y, and the cohomology class of ρδ\rho_{\delta} in Hcn​(Nδ,𝐑)H^{n}_{c}(N_{\delta},{{\bf R}}) is the same as of [d​i​a​g][diag].

We define Rδ:D′​(Y)→Ω∗​(Y)R_{\delta}:D^{\prime}(Y)\to\Omega^{\ast}(Y) as the integral operator given by the kernel ρδ\rho_{\delta}.

Lemma 3

1) The operator RδR_{\delta} is a homomorphism of complexes.

2) If Z1,Z2∈YZ_{1},Z_{2}\in Y are two oriented submanifolds of finite volume such that they intersect transversally at finitely many points, and d​i​m​Z1+d​i​m​Z2=d​i​m​Ydim\,Z_{1}+dim\,Z_{2}=dim\,Y, Z¯1∩Z¯2=Z1∩Z2\overline{Z}_{1}\cap\overline{Z}_{2}={Z}_{1}\cap{Z}_{2}, then for sufficiently small δ\delta one has:

∫YRδ​([Z1])∧Rδ​([Z2])=d​e​g​(Z1∩Z2)∈𝐙\int_{Y}R_{\delta}([Z_{1}])\wedge R_{\delta}([Z_{2}])=deg(Z_{1}\cap Z_{2})\in{\bf Z}

3) There exists a linear operator hδ:Ω∗​(Y)→Ω∗​(Y)h_{\delta}:\Omega^{\ast}(Y)\to\Omega^{\ast}(Y) such that its kernel has support in NδN_{\delta}, the wave front W​F​(hδ)WF(h_{\delta}) is the conormal bundle of d​i​a​g⊂Y×Ydiag\subset Y\times Y, and

dhδ+hδd=id−(Rδ)|Ω∗(Y).dh_{\delta}+h_{\delta}d=id-(R_{\delta})_{|\Omega^{\ast}(Y)}.

Proof . Part 1) follows from the fact that ρδ\rho_{\delta} is a closed current. Part 2) follows from the fact that RδR_{\delta} changes the supports of Zi,i=1,2Z_{i},i=1,2 by O⁡(δ)O(\delta). To prove part 3) one observes that the operators i​did and (Rδ)|Ω∗(Y)(R_{\delta})_{|\Omega^{\ast}(Y)} preserve the space of smooth forms Ω∗​(Y)\Omega^{\ast}(Y), and ρδ\rho_{\delta} is cohomologous to [d​i​a​g][diag]. ■\blacksquare

Let x,y∈C​r​(f)x,y\in Cr(f) be two critical points of the same Morse index. Then d​e​g​(Sx∩Uy)=δx​ydeg(S_{x}\cap U_{y})=\delta_{xy} (the Kronecker symbol). By the part 2) of the Lemma, for sufficiently small δ\delta we obtain the identity

∫YRδ​([Sx])∧Rδ​([Uy])=δx​y\int_{Y}R_{\delta}([S_{x}])\wedge R_{\delta}([U_{y}])=\delta_{xy}

This implies the following result.

Proposition 6

Let us define for a sufficiently small δ\delta a linear operator D′​(Y)→Ω∗​(Y)D^{\prime}(Y)\to\Omega^{\ast}(Y) by the formula Πδ​(α)=∑x∈C​r​(f)(∫Yα∧Rδ​([Sx]))⋅Rδ​([Ux]).\Pi_{\delta}(\alpha)=\sum_{x\in Cr(f)}(\int_{Y}\alpha\wedge R_{\delta}([S_{x}]))\cdot R_{\delta}([U_{x}]).

Then

1) Πδ2​(α)=Πδ​(α)\Pi_{\delta}^{2}(\alpha)=\Pi_{\delta}(\alpha) if α∈Ω∗​(Y)\alpha\in\Omega^{\ast}(Y), and Πδ​d=d​Πδ.\Pi_{\delta}d=d\Pi_{\delta}.

2) The image Πδ​(M∗​(Y,f))\Pi_{\delta}(M^{\ast}(Y,f)) is a subcomplex in Ω∗​(Y)\Omega^{\ast}(Y) which is canonically isomorphic to the Morse complex M∗​(Y,f)M^{\ast}(Y,f).

We define a homotopy operator Hδ:Ω∗​(Y)→Ω∗​(Y)​[−1]H_{\delta}:\Omega^{\ast}(Y)\to\Omega^{\ast}(Y)[-1] as an integral operator given by the kernel (Rδ⊠Rδ)​[Z∞]+(hδ⊠hδ)​([d​i​a​g]).(R_{\delta}\boxtimes R_{\delta})[Z_{\infty}]+(h_{\delta}\boxtimes h_{\delta})([diag]). (The last summand is well-defined because of the condition on the wave front of hδh_{\delta}). It is easy to check that the following identity holds:

i​d−Πδ=d​Hδ+Hδ​d.id-\Pi_{\delta}=dH_{\delta}+H_{\delta}d.

Thus we have a family of homotopies and projectors parametrized by δ\delta.

Remark 17

One can define the projector Πδ\Pi_{\delta} using another canonical element ∑x∈C​r​(f)[Sx]⊗Rδ​([Ux])\sum_{x\in Cr(f)}[S_{x}]\otimes R_{\delta}([U_{x}]), instead of ∑x∈C​r​(f)Rδ​([Sx])⊗Rδ​([Ux])\sum_{x\in Cr(f)}R_{\delta}([S_{x}])\otimes R_{\delta}([U_{x}]), as we did. The above Proposition holds for the new canonical element as well.

There is a version of the previous construction, which will be useful in the next subsection. Namely, we start with a differential n+1n+1-form ρ\rho on Y×Y×(0,1)Y\times Y\times(0,1) such that for the support of s​u​p​p​(ρ)supp(\rho) belongs to ⊔δ>0(Nδ,δ)\sqcup_{\delta>0}(N_{\delta},\delta) for all sufficiently small δ∈(0,1)\delta\in(0,1), and ρ\rho defines the same cohomology class in Hcn​(Y×Y×(0,1))H_{c}^{n}(Y\times Y\times(0,1)) as [d​i​a​g]×(0,1)[diag]\times(0,1).

Let us consider now the spaces Ω0∗​(Y):=lim→δ→0⁡Ω∗​(Y×(0,δ))\Omega_{0}^{\ast}(Y):=\varinjlim_{\delta\to 0}\Omega^{\ast}(Y\times(0,\delta)) and D0′​(Y):=lim→δ→0⁡Ω∗​(0,δ)​⊗^​D′​(Y).D_{0}^{\prime}(Y):=\varinjlim_{\delta\to 0}\Omega^{\ast}(0,\delta)\widehat{\otimes}D^{\prime}(Y). It is easy to see that both complexes Ω0∗​(Y)\Omega_{0}^{\ast}(Y) and D0′​(Y)D_{0}^{\prime}(Y) are quasi-isomorphic to Ω∗​(Y)\Omega^{\ast}(Y) .

We define a linear operator R:D0′​(Y)→Ω0∗​(Y)R:D_{0}^{\prime}(Y)\to\Omega_{0}^{\ast}(Y) similarly to the definition of RδR_{\delta}. Then the Lemma and the Proposition hold with obvious changes. We will denote the corresponding objects by the same letters as before, skipping the subscript δ\delta (like HH for the homotopy and Π\Pi for the projector). Morally, they are obtained from the old objects by extending them as differential forms “in the direction of δ\delta”.

Original mathematics by the credited authors. Source-backed reader collection; mathematical self-containment is not assessed.