arXiv reader

Cyclic theory for commutative differential graded algebras and s-cohomology

Dan Burghelea

arXiv
0905.1489v1 · PDF · 10 May 2009
Categories
math.AT, math.KT
Licence
CC BY 4.0 The site shows the text.

Abstract

In this paper one considers three homotopy functors on the category of manifolds, hH^\ast, cH^\ast, sH^\ast, and parallel them with other three homotopy functors on the category of connected commutative differential graded algebras, HH^\ast, CH^\ast, SH^\ast. If P is a smooth 1-connected manifold and the algebra is the de-Rham algebra of P the two pairs of functors agree but in general do not. The functors HH^\ast and CH^\ast can be also derived as Hochschild resp. cyclic homology of commutative differential graded algebra, but this is not the way they are introduced here. The third SH^\ast , although inspired from negative cyclic homology, can not be identified with any sort of cyclic homology of any algebra. The functor sH^\ast might play some role in topology. Important tools in the construction of the functors HH^\ast, CH^\astand SH^\ast , in addition to the linear algebra suggested by cyclic theory, are Sullivan minimal model theorem and the "free loop" construction described in this paper.

Items 19

Proposition

Proposition 1

N8YW

Let (C, \delta, \beta) be a mixed cochain complex.

1.PH^r = \underset {\underset S{\rightarrow}} \lim ^+H^{r+2k}_\beta, where S^{k+2r}: ^+H^{k+2r}_\beta \to ^+H^{k+2r+2}_\beta is induced by the inclusion ^+ \mathcal C^{\ast}_\beta \to ^+\mathcal C^{\ast+2}_\beta.

2. The following is an exact sequence

&0 &S ’ ^+H^_r-1 +2k &P H^r , & S ^+H^_r+2k& 0, with S_{k+2r}: ^+H_{k+2r}^\delta \to ^+H_{k+2r-2}^\delta induced by the projection ^+\mathcal H^\delta_\ast\to ^+\mathcal H^\delta_{\ast-2}.

Let f^\ast:(C^\ast_1, \delta^\ast _1, \beta^1_\ast) \to (C^\ast_2, \delta^\ast _2, \beta^2_\ast) be a morphism of mixed complexes.

3. If H^\ast(f) is an isomorphism then so is ^+H^\ast_\beta (f) and PH^\ast (f).

4. If H_\ast (f) is an isomorphism then so is ^+H^\delta_\ast (f) and \mathbb PH^\ast(f).

5. If H^\ast(f) and H_\ast(f) are both isomorphisms then, in addition to the conclusions in (3) and (4), ^-H^\ast_\beta(f) is an isomorphism.

Used by Proof
Definition

Definition 1

J22C

    2em A commutative graded algebra abbreviated CGA, is an associative unital augmentable graded algebra \mathcal A^\ast, (the augmentation is not part of the data) which is commutative in the graded sense, i.e.

    a_1\cdot a_2= (-1)^{r_1 r_2}a_2\cdot a_1,\ \ a_i\in \mathcal A^{r_i}, i=1,2.

    2em An exterior differential d^\ast_\mathcal A: \mathcal A^\ast \to \mathcal A^{\ast+1}, is a degree +1-linear map which satisfies

    d(a_1\cdot a_2)= d(a_1)\cdot a_2 +(-1)^{r_1} a_1\cdot d(a_2), a_1\in \mathcal A^{r_1} ,\ \ d^{*+1}_{\mathcal A} d^{*}_{\mathcal A}=0.

    2em An interior differential \beta^\mathcal A_\ast : \mathcal A^\ast\to \mathcal A^{\ast-1} is a degree -1 linear map which satisfies

    \beta(a_1\cdot a_2)= \beta(a_1)\cdot a_2 +(-1)^{r_1} a_1\cdot \beta(a_2), a_1\in \mathcal A^{r_1}, \ \ \beta_{\ast-1}^{\mathcal A} \beta_{\ast}^{\mathcal A}=0.

    2em The exterior and interior differentials d^\ast and \beta_\ast are compatible if

    d^{\ast-1}\cdot \beta_\ast +\beta_{\ast+1}\cdot d^\ast=0.

    2em A pair (\mathcal A^\ast, d^\ast),\ \mathcal A^\ast a CGA and d^\ast exterior differential, is called CDGA and a triple (\mathcal A^\ast, d^\ast, \beta_\ast),\ \mathcal A^\ast a CGA, d^\ast exterior differential and \beta_\ast interior differential, with d^\ast and \beta_\ast compatible, is called a mixed CDGA.

Definition

Definition 2

BQSD

A morphism of CDGA’s f, with H^k(f) isomorphism for any k, is called a quasi isomorphism.

The CDGA (\mathcal A, d_{\mathcal A}) is called homologically connected if H^0(\mathcal A, d_{\mathcal A}) =\kappa and homologically 1-connected if homologically connected and H^1(\mathcal A, d_{\mathcal A})=0.

Definition

Definition 3

66NQ

1. The CDGA (\mathcal A, d) is called free if \mathcal A= \Lambda [V], where V= \sum_{i\geq 0}V^i is a graded vector space and \Lambda[V] denotes the free commutative graded algebra generated by V. If in addition V^0=0 is called free connected commutative differential graded algebra, abbreviated fc-CDGA.

2. The CDGA (\mathcal A, d) is called minimal if it is a fc-CDGA and in addition

i. d(V)\subset \Lambda^+[V]\cdot \Lambda^+[V], with \Lambda^+[V] the ideal generated by V,

ii. V^1= \oplus_{\alpha\in I} V_\alpha with I a well ordered set and d(V_\beta)\subset \Lambda[\oplus _{\alpha <\beta} V_\alpha] (the set I and its order are not part of the data)

Observation

Observation 1

OTP1

If (\Lambda[V], d_V) is minimal and 1-connected, then V^1=0 and, for v\in V^i, d_V (v)is a linear combination of products of elements v_j\in V^j with j<i. In particular for v\in V^2 one has dv=0.

Theorem

Theorem 1

R22M

1 (D. Sullivan)

1. A quasi isomorphism between two minimal CDGA’s is an isomorphism.

2. For any homologically connected CDGA, (\mathcal A, d_{\mathcal A}), there exists quasi isomorphisms \theta: (\Lambda[ V], d_V)\to (\mathcal A, d_{\mathcal A}) with (\Lambda[ V], d_V) minimal. Such \theta will be called minimal model of (\mathcal A, d_\mathcal A).

3. Given a morphism f:(\mathcal A, d_{\mathcal A})\to (\mathcal B, d_{\mathcal B}) and the minimal models \theta_A: (\Lambda[V_A], d_{V_A})\to (A,d_A) and \theta_B: (\Lambda[V_B], d_{V_B})\to (B,d_B), there exists morphisms f' : (\Lambda [V_A], d_{V_A}) \to (\Lambda[V_B], d_{V_B}) such that f \cdot \theta_A and \theta_B\cdot f' are homotopic; moreover any two such (f')’s are homotopic.

Definition

Definition 4

M6EI

A linear representation \rho :S^1\times V\to V on the good Frechet space is good if the following conditions hold:

a. V^f, the fixed point set, is a good Frechet space,

b. The map r^\ast: \Omega^\ast (V) \to \Omega^\ast (V^f) induced by the inclusion is surjective,

c. (\Omega^\ast_X (V, V^f), i^X_\ast) with (\Omega^\ast_X (V, V^f) = \ker r^\ast is acyclic.

Proposition

Proposition 2

LA0O

1. Any representation on a finite dimensional vector space is good.

2. If V is a good Frechet space then the regular representation \rho: S^1 \times C^\infty (S^1, V)\to C^\infty (S^1, V), with C^\infty (S^1, V), the Frechet space of smooth functions, is good.

Used by Proof
Definition

Definition 5

2U62

A smooth action \mu: S^1\times M\to M is good if its linearization at any fixed point is a good representation.

Proposition

Proposition 3

MF84

If \tilde M= (M,\mu) is a smooth S^1-manifold and X is the associated vector field, then:

1. H^\ast_X(M)= H^\ast (M),

2. ^+ H^\ast_X(M)= H^\ast_{S^1}(\tilde M) and S:H^\ast(M)\to H^{\ast +2}(M) identifies to the multiplication with u\in H^2_{S^1}(pt), the generator of the equivariant cohomology of one point space,

3. PH^\ast_X(M)= \underset {\underset {S} {\rightarrow}} {\lim} H^{\ast+2k}_{S^1}(\tilde M).

If the action is good then:

4. \mathbb PH^\ast_X(M)= K^\ast (F)\ where

\begin{aligned} K^r(F)= &\prod_k H^{2k}(F)\ \text {if}\ \ r \ \ even\\ K^r(F)= &\prod_k H^{2k+1}(F)\ \text {if}\ \ r \ \ odd. \end{aligned}

If M is a closed of n-dimension manifold then :

5. ^- H^\ast_X(M)= H^{S^1}_{n-1-\ast} (\tilde M, \mathcal O_M) with H^{S^1}_\ast (\tilde M, \mathcal O_M) the equivariant homology with coefficients in the orientation bundle ( Recall that H^{S^1}_{\ast}(M,\mathcal O_M)= H_\ast ( M// S^1, \mathbb O_M) where M//S^1 is the homotopy quotient of this action. This equivariant homology can be derived from invariant currents in the same way as equivariant cohomology from invariant forms, cf. cite AB. The complex of invariant currents (with coefficients in orientation bundle) contains the complex (\Omega^{n-\ast} _X(M, \mathcal O_M), \partial _{n-\ast}) as a quasi isomorphic sub complex.) of M.

Proof of Proposition 3

1. The verification is standard since S^1 is compact and connected; one construct av^\ast: (\Omega^\ast(M), d^\ast) \to (\Omega_X^\ast, d^\ast) by S^1- averaging using the compacity of S^1. The homomorphism induced in cohomology by av^\ast is obviously surjective. To check it is injective one has to show that any closed k differential form \omega which becomes exact after applying av is already exact, precisely \int_c\omega=0 for any smooth k- cycle c. Indeed, since the connectivity of S^1 implies \int_c\omega= \int_{\mu(-\theta, c)}\omega , \ \theta \in S^1 one has:

\begin{aligned} \int_c\omega= 1/2\pi \int_{S^1}(\int_c\omega)d\theta= \\ 1/2\pi \int_{S^1}(\int_{\mu(-\theta, c)} \omega)d\theta =1/2\pi \int_{S^1}(\int_c \mu_\theta^\ast (\omega))d\theta= \\ \int_c (1/2\pi \int_{S^1} \mu_\theta^\ast (\omega)d\theta)= \int_c av^\ast(\omega)=0. \end{aligned}

Here \mu_\theta denotes the diffeomorphism \mu(\theta, \cdot ): M\to M.

2. Looking at the definition in section r@SMCSMC one recognizes one of the most familiar definition of equivariant cohomology using invariant differential forms cf. cite AB.

3. The proof is a straightforward consequence of Proposition r@P1P1 in section r@SMCSMC and (2) above.

4. Let F be the smooth sub manifold of the fixed points of \mu. Clearly r^\ast : (\Omega^\ast_X(M), d^\ast, i^X_\ast) \to (\Omega^\ast (F), d^\ast_F, 0) is a morphism of mixed CDGA, hence of mixed complexes. If the smooth action is good then the above morphism induces an isomorphism in homology H_\ast (\Omega^\ast_X, i^X_\ast)\to H_\ast (\Omega^\ast (F), 0). To check this we have to show that (\ker r^\ast, i^X_\ast), with \ker r^\ast:= \{\omega\in \Omega_X ^\ast(M) | \omega|_F=0\}, is acyclic. This follows (by S^1-average) from the acyclicity of the chain complex (\Omega^\ast (M,F), i^X_\ast) which in turn can be derived using the linearity w.r. to functions of of i^X. Indeed, using a "partition of unity" argument it suffices to verify this acyclicity locally. For points outside F the acyclicity follows from the acyclicity of the complex

\cdots \Lambda^{\ast-1} (V) \overset {i^e}\to \Lambda^\ast(V)\overset{i^e}\to \Lambda ^{\ast-1}(V)\to \cdots , where V is a Frechet space , \Lambda^k (F) the space of skew symmetric k-linear maps from V to \kappa= \mathbb R, \mathbb C and e\in V\setminus 0. For points x\in F this follows from the fact that the linearization of the action at x is a good representation, as stated in Proposition r@P2P2.

5. If \tilde M is a finite dimensional smooth S^1-manifold we can equip M with an invariant Riemannian metric g and consider \star:\Omega^\ast(M)\to \Omega^{n-\ast}(M; \mathcal O_M) the Hodge star operator. Denote by \omega\in \Omega^1 (M) the 1-form corresponding to X w.r. to the metric g, by e_\omega:\Omega^\ast(M;\mathcal O_M)\to \Omega^{\ast +1}(M;\mathcal O_M) the exterior multiplication with \omega and by \partial_\ast: \Omega^\ast(M;\mathcal O_M)\to \Omega^{\ast-1}(M;\mathcal O_M) the formal adjoint of d^{\ast-1} w.r. to g i.e. \partial _\ast = \pm \star \cdot\ d ^{n-\ast}\cdot \ \star^{-1} . Note that e_\omega= \pm \star \cdot \ i^X \cdot \star^{-1} . All these operators leave \Omega_{X} invariant since g is invariant. Clearly (\Omega ^\ast_{X}(M;\mathcal O_M), e^\ast _\omega, \partial _\ast ) is a mixed cochain complex and we have

^-H^\ast _{i^X}(\Omega_X (M), d, i^X)= ^+H^\partial_{n-\ast} (\Omega_X (M;\mathcal O_M), e^\omega,\partial ).

The equivariant homology of \tilde M with coefficients in the orientation bundle can be calculated from the complex of invariant currents which, if M closed, contains the complex (\Omega^{n-\ast} _X(M, \mathcal O_M), \partial _{n-\ast}) as a quasi isomorphic sub complex. As a consequence we have

\begin{aligned} H_{n-\ast}(\Omega_X(M;\mathcal O_M),\partial)= &H_\ast(M;\mathcal O_M)\\ ^+H^{e_\omega}_{n-\ast}(\Omega_X(M;\mathcal O_M), e_\omega,\partial)= &H_\ast^{S^1}(M;\mathcal O_M) \end{aligned}

(cf. section r@SMCSMC for notations).

Theorem

Theorem 2

QBMQ

If P is a 1-connected smooth manifold then we have the following short exact sequence:

0 \to \overline K^r(P)\otimes \kappa \overset {\mathbb J^r}\rightarrow {sH}^r (P) \overset {\mathbb B^r}\rightarrow \overline cH^{r-1}(P)\to 0

where \kappa= \mathbb R or \mathbb C.

Observation

Observation 2

B9K3

The vector space K^\ast(P) can be identified via the Chern character to the Atiyah–Hirtzebruch (complex) K-theory tensored with the field \kappa= \mathbb R \text{or} \ \mathbb C, depending of what sort of differential forms one consider (real or complex valued). When P is 1-connected \overline cH^\ast (P) identifies to \text{Hom}(\tilde A_{\ast}(P), k) where \tilde A_{\ast}(P) denotes the reduced Waldhaussen algebraic K-theory(often referred to as A- theory.), cf cite B2. From this perspective sH^\ast unifies topological (Atiyah–Hirtzebruch) K-theory and Waldhaussen algebraic K-theory.

Observation

Observation 3

UXOW

In view of the definition of ^-H^\ast_\beta(C^\ast, \delta^\ast, \beta_\ast), cf. section r@SMCSMC, observe that sH^\ast(P) is represented by infinite sequences (sH^\ast (P) is the cohomology of the cochain complex (^-C^\ast, ^-D^\ast) with ^- C^r= \prod _{k\geq 0} \Omega^{r+2k}_{inv} (P^{S^1}) and

^-D^r(\cdots, \omega_{r+2}, \omega_{r})= (\cdots, (i^X \omega_{r+2} + d \omega_r)), cf. section r@SMCSMC.), rather than eventually finite sequences of invariant differential forms on P^{S^1}. If instead of “infinite sequences" we would have considered “eventually finite sequences” the outcome would have been different for infinite dimensional manifolds.

The difference between “infinite sequences" and "eventually finite sequences" exists only for infinite dimensional manifolds which P^{S^1} always is.

Theorem

Theorem 3

5346

Let P be a 1-connected manifold.

1. All eigenvalues of the endomorphisms h\Psi_k and c\Psi_k are k^r, r=0,1, 2\cdots, and the eigenspaces corresponding to k^r are independent of k provided k\geq 2.

2. Denotes these eigenspaces by hH^\ast(M)(r) and cH^\ast (M)(r). Then

hH^\ast(0)= H^\ast(X;\kappa), cH^{\ast}(0)= H^{\ast+1}(X;\kappa), and

hH^r(p)= cH^r(p)= 0, p\geq r+1.

3. If \sum_i \dim \pi_i(P)\otimes \kappa < \infty, \kappa the field of real or complex numbers and \sum_i \dim (H^i(P) )<\infty then for any r\geq 0 one has

\sum_i \dim hH^i(P) (r) < \infty, \ \ \sum_i\dim cH^i(P)(r) < \infty.
Observation

Observation 4

PDH9

1. \text({Im}(i^V) ,\delta_V, 0) is a mixed sub complex of (\Lambda^+[V \oplus \overline V],\delta_V, i^V)\subset (\Lambda[V\oplus \overline V],\delta_V, i^V)

2. \Psi_k, k= -1, 1, 2,\cdots leave (\Lambda^+[V\oplus \overline V],\delta_V, i^V) and (\text{Im}(i^V) ,\delta) invariant and have k^r r=0,1,2,\cdots as eigenvalues. These are all eigenvalues.

For k \geq 2 the eigenspace of \Psi_r: \Lambda[V\oplus \overline V]\to \Lambda [V\oplus \overline V] corresponding to the eigenvalue k^r is exactly \Lambda[V]\otimes \overline V^{\otimes r}, resp. \Lambda^+[V\oplus \overline V]\cap \Lambda^[V]\otimes \overline V^{\otimes r}, resp. \text{Im}(i^V)(r)= \text{Im}(i^V)\cap \Lambda [V] \otimes \overline V^{\otimes r}, hence independent of k. Each such eigenspace is \delta_V-invariant.

3. The mixed complex (\Lambda^+[V\oplus \overline V],\delta_V, i^V) is i^V-acyclic,

4. We have the decomposition

(\Lambda [V\oplus \overline V], \delta_V)= \bigoplus_{r\geq 1} (\Lambda[V]\otimes \overline V^{\otimes r}, \delta_V)

and the analogous decomposition for (\Lambda^+ [V\oplus \overline V], \delta_V) and (\text{Im}(i^V), \delta_V) referred from now on as the weight decompositions.

Used by Proof
Observation

Observation 5

46FX

1. (\Lambda[V]\otimes F_p(\overline V^{\otimes r}), \delta_V) is a sub complex of (\Lambda[V]\otimes \overline V^{\otimes r}, \delta_V).

2.If (\Lambda[V], d_V) is minimal and one connected then, by Observation r@O0O0,

\delta (F_p(\overline V^{\otimes r})) \subset \Lambda[V]\otimes F_{p-1}(\overline V^{\otimes r}) and then

( {\Lambda[V]\otimes F_p(\overline V^{\otimes r})}/{ \Lambda[V]\otimes F_{p-1}(\overline V^{\otimes r})} \ , \delta_V)= (\Lambda[V], d_V)\otimes {F_p(\overline V^{\otimes r})}/ {F_{p-1}(\overline V^{\otimes r})}.

2. \sum_p \dim (F_p(\overline V^{\otimes r}) / F_{p-1}(\overline V^{\otimes r}))= \dim (\overline V^{\otimes r})= (\dim V)^r.

Uses Observation 1 · Used by Proof
Theorem

Theorem 4

QQSH

Let (\mathcal A, d_\mathcal A) be a connected CDGA.

1. All eigenvalues of the endomorphisms H\Psi_k and C\Psi_k are k^r, r=0,1, 2\cdots, and their eigenspaces are independent of k provided k\geq 2. One denotes them by HH(\mathcal A, d_\mathcal A)(r), and CH(\mathcal A, d_\mathcal A)(r).

2.

\begin{aligned} HH^\ast(\mathcal A, d_\mathcal A)(0)=&H^\ast(\mathcal A, d_\mathcal A), \\ CH^\ast(\mathcal A, d_\mathcal A)(0)= &H^{\ast+1}(\mathcal A, d_\mathcal A)\\ HH^r(\mathcal A, d_\mathcal A)(p)= &CH^r(\mathcal A, d_\mathcal A)(p)=0, \ \ p\geq r+1 \end{aligned}

3.Suppose (\mathcal A, d_\mathcal A) is 1-connected with minimal model (\Lambda[V], d_V). If \sum_i\dim V^i< \infty and \sum_i \dim H^i(\mathcal A, d_\mathcal A) <\infty then for any r\geq 0 one has

\sum_i \dim HH^i(\mathcal A, d_\mathcal A)(r) < \infty, \ \ \sum_i\dim CH^i(\mathcal A, d_\mathcal A)(r) < \infty.
Proof of Theorem 4

It suffices to check the statements for (\mathcal A, d_\mathcal A)=(\Lambda[V],d_V) minimal. Items 1) and 2) are immediate consequences of Observation r@O3O3.

Item 3) follows from Observation r@O4O4. Indeed for a fixed r one has

\begin{aligned} \sum_{i} \dim H^i (\Lambda[V]\otimes \overline V^{\otimes r}, \delta_V)) \leq \\ \sum_{i, p} \dim H^i ({\Lambda[V]\otimes F_p(\overline V^{\otimes r}}/{ \Lambda[V]\otimes F_{p-1}(\overline V^{\otimes r})}, \ \delta_V)= \\ (\dim V)^r \cdot \sum_{i} \dim H^i (\Lambda[V], d_V) \end{aligned}

Source and licence

The text on this page is from arXiv:0905.1489v1 by Dan Burghelea, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/0905/0905.1489.jsonl.