This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Asymptotic pseudodifferential calculus and the rescaled bundle

Xiaoman Chen and Zelin Yi
Abstract

By following a groupoid approach to pseudodifferential calculus developed by Van erp and Yuncken, we study the parallel theory on the rescaled bundle and show that the rescaled bundle gives a geometric characterization to asymptotic pseudodifferential calculus on spinor bundles by Block and Fox.

1 Introduction

Pseudodifferential operators have pseudo-local property, that is, their Schwartz kernels are smooth outside the diagonal. As a consequence, any pseudodifferential operator can be decomposed as a properly supported pseudodifferential operator plus a smoothing operator. Ordinary principal symbol calculus, which is captured by the construction of tangent groupoid, focus on the quotient space Ψm(M)/Ψ(M)\Psi^{m}(M)/\Psi^{-\infty}(M) where one cannot tell the difference between smoothing operator and the zero operator. However, as far as local index theory is concerned, it is precisely smoothing operators that carry index information. More precisely, according to the Mckean-Singer formula[MS67], the topological index can be calculated by supertrace of the smoothing operator given by the heat kernel of the Dirac operator. Therefore one need a full symbol calculus to explore the space of smoothing operators. Along this line, Widom have developed an asymptotic symbolic calculus where a smooth family of full symbols a(x,ξ,t)a(x,\xi,t) of order mm is studied from the view-point of asymptotic expansion

(1.1) a(x,ξ,t)a0(x,ξ)+ta1(x,ξ)+t2a2(x,ξ)+.a(x,\xi,t)\sim a_{0}(x,\xi)+ta_{1}(x,\xi)+t^{2}a_{2}(x,\xi)+\cdots.

where ai(x,ξ)a_{i}(x,\xi) is a symbol of order mim-i and the asymptotic expansion means for all NN, the quantity

tN(a(x,ξ,t)k=0N1tkak(x,ξ))t^{-N}\left(a(x,\xi,t)-\sum_{k=0}^{N-1}t^{k}a_{k}(x,\xi)\right)

converges to zero in the space of symbols of order mNm-N as t0t\to 0. It has the advantage that its calculus is easier than that of full symbols and, at the same time, useful aspect of smoothing operators, for example the asymptotic expansion of the operator trace at t=0t=0, are preserved.

To adept this idea into the realm of local index theory, Block and Fox[BF90] developed the asymptotic pseudodifferential calculus on spinor bundles over compact spin manifold and use it, together with the JLO formula, to calculate Connes’ cyclic cocycle. In this paper, we recover the calculus from a geometric point-view by using a groupoid approach to pseudodifferential calculus by Van erp and Yuncken[EY17] in the context of rescaled bundle[HY19].

Let MM be an even dimensional spin manifold with spinor bundle SMS\to M. The rescaled bundle 𝕊\mathbb{S} is a vector bundle over the tangent groupoid 𝕋M\mathbb{T}M

(1.2) πTM\textstyle{\pi^{\ast}\wedge T^{\ast}M\ignorespaces\ignorespaces\ignorespaces\ignorespaces}SS\textstyle{S^{\ast}\boxtimes S\ignorespaces\ignorespaces\ignorespaces\ignorespaces}TM×{0}\textstyle{TM\times\{0\}}\textstyle{\sqcup}M×M×.\textstyle{M\times M\times\mathbb{R}^{\ast}.}

whose restriction to TM×{0}TM\times\{0\} is the pullback of the bundle of exterior algebras and restriction to M×M×M\times M\times\mathbb{R}^{\ast} is the pullback of the tensor product SSS^{\ast}\boxtimes S. Moreover, there is an open neighborhood 𝕌\mathbb{U} of TM×{0}TM\times\{0\} inside the tangent groupoid which is homeomorphic to an open neighborhood 𝕌~\widetilde{\mathbb{U}} of TM×{0}TM\times\{0\} inside TM×TM\times\mathbb{R}. Let ρ:TM×M\rho:TM\times\mathbb{R}\to M be the bundle projection, there is a natural isomorphism between the restricted rescaled bundle 𝕊|𝕌\mathbb{S}|_{\mathbb{U}} and the bundle of exterior algebra ρTM|𝕌~\rho^{\ast}\wedge T^{\ast}M|_{\widetilde{\mathbb{U}}}. This fact together with the smooth structure of SSS^{\ast}\boxtimes S over M×M×M\times M\times\mathbb{R}^{\ast} completely determine the smooth structure of the rescaled bundle.

The rescaled bundle also carries a multiplicative structure which is given by the smoothly varying maps

(1.3) 𝕊γ𝕊η𝕊γη\mathbb{S}_{\gamma}\otimes\mathbb{S}_{\eta}\to\mathbb{S}_{\gamma\circ\eta}

where (γ,η)(\gamma,\eta) is a composable pair of elements in the tangent groupoid and γη\gamma\circ\eta is their groupoid multiplication. When γ=(x,X,0)\gamma=(x,X,0) and η=(x,Y,0)\eta=(x,Y,0) come from TM×{0}TM\times\{0\} part of the tangent groupoid, the multiplication map (1.3) is explicitly computable as

(1.4) TxMTxMαβexp(12κ(X,Y))αβ\wedge T^{\ast}_{x}M\otimes\wedge T^{\ast}_{x}M\ni\alpha\otimes\beta\mapsto\exp\left(-\frac{1}{2}\kappa(X,Y)\right)\wedge\alpha\wedge\beta

where κ(X,Y)\kappa(X,Y) is a differential 2-form given by the symbol of curvature of the spinor bundle.

Following [LMV17] and [EY17], the space of properly supported rr-fibered distribution on the rescaled bundle r(𝕋M,𝕊)\mathcal{E}^{\prime}_{r}(\mathbb{T}M,\mathbb{S}) is defined to be the set of continuous C(M×)C^{\infty}(M\times\mathbb{R})-module maps C(𝕋M,𝕊)C(M×)C^{\infty}(\mathbb{T}M,\mathbb{S})\to C^{\infty}(M\times\mathbb{R}) and we shall consider the subspace Ψm(𝕋M,𝕊)\Psi^{m}(\mathbb{T}M,\mathbb{S}) that consists of r(𝕋M,𝕊)\mathbb{P}\in\mathcal{E}^{\prime}_{r}(\mathbb{T}M,\mathbb{S}) that satisfy the essentially homogeneous condition

(1.5) αλ,λmCp(𝕋M,𝕊)\alpha_{\lambda,\ast}\mathbb{P}-\lambda^{m}\mathbb{P}\in C_{p}^{\infty}(\mathbb{T}M,\mathbb{S})

where αλ\alpha_{\lambda} is a smooth action on the tangent groupoid that sends (x,y,t)(x,y,t) to (x,y,λt)(x,y,\lambda t) for t0t\neq 0 and (x,Y,0)(x,Y,0) to (x,λ1Y,0)(x,\lambda^{-1}Y,0). In this paper, we shall focus on the Taylor’s expansion of Ψm(𝕋M,𝕊)\mathbb{P}\in\Psi^{m}(\mathbb{T}M,\mathbb{S}) at t=0t=0. According to the smooth structure of the rescaled bundle, a distribution Ψm(𝕋M,𝕊)\mathbb{P}\in\Psi^{m}(\mathbb{T}M,\mathbb{S}) may be restricted to an open neighborhood of TM×{0}TM\times\{0\} and full symbol of \mathbb{P} is defined by its Fourier transformation.

Although the full symbol contains more information than the principal symbol, it is difficult to do calculation with (for example, its composition formula is complicated). In order to preserve useful aspects of smoothing operator, instead of the space of full symbols we shall study the space of Taylor’s expansion at t=0t=0

(1.6) a(x,ξ,t)a(x,ξ,0)+tta(x,ξ,0)+t22t2a(x,ξ,0)+a(x,\xi,t)\sim a(x,\xi,0)+t\partial_{t}a(x,\xi,0)+\frac{t^{2}}{2}\partial^{2}_{t}a(x,\xi,0)+\cdots

It has two advantages

  • the multiplicative structure (1.3) induces an algebra structure on Ψm(𝕋M,𝕊)\Psi^{m}(\mathbb{T}M,\mathbb{S}) which in turn induces an algebra structure on the space of Taylor’s expansion of full symbols. The crucial point is that the multiplication formula of Taylor’s expansion is explicitly computable and a lot easier than that of full symbols;

  • If a(x,ξ,t)a(x,\xi,t) is a full symbol of order less than or equal to n-n, than the supertrace of the corresponding pseudodifferential kernel \mathbb{P} has asymptotic expansion

    Str()iti(2i)n/2(2π)nTMai(x,ξ)𝑑ξ𝑑x\operatorname{Str}(\mathbb{P})\sim\sum_{i}t^{i}\cdot\left(\frac{2}{i}\right)^{n/2}(2\pi)^{-n}\int_{T^{\ast}M}a_{i}(x,\xi)d\xi dx

However, this space is restrictive in two ways:

  • it represents only properly supported distributions;

  • all ai(x,ξ)a_{i}(x,\xi)’s in the expansion (1.6) are homogeneous modulo Schwartz functions.

There is an important class of pseudodifferential operators: the heat kernel that falls out of this category. We shall enlarge the space by looking at the space of symbols SmS^{m} which is defined to be the subspace of C(TM×,ρTM)C^{\infty}(T^{\ast}M\times\mathbb{R},\rho^{\ast}\wedge T^{\ast}M) satisfy the classical symbol estimate

(1.7) |xαξβtγa(x,ξ,t)|C(1+|ξ|+t)m|β|,\left|\partial_{x}^{\alpha}\partial_{\xi}^{\beta}\partial_{t}^{\gamma}a(x,\xi,t)\right|\leq C\cdot(1+|\xi|+t)^{m-|\beta|},

for all α,β,γ\alpha,\beta,\gamma, here ρ:TM×M\rho:T^{\ast}M\times\mathbb{R}\to M is the bundle projection. The space of Taylor’s expansion associate with this extended symbol space retains the two advantages. Moreover, the heat kernel, which can be expressed as the fundamental solution to the differential equation

τ+t2D2=0,\frac{\partial}{\partial\tau}+t^{2}D^{2}=0,

is captured by this class of symbols. By first passing to distributions level, second to the full symbol level and third to the Taylor’s expansion level of the differential equation, combine with the fact that multiplication formula of Taylor’s expansion is explicitly computable, the differential equations are simplified and the solutions are explicitly computable. Under this light, the asymptotic expansion of the heat kernel eτt2D2e^{-\tau t^{2}D^{2}} can be calculated and the Mckean-Singer formula guarantee that the supertrace of the leading term is precisely the topological index. As a corollary, we obtain that the heat kernel eτt2D2e^{-\tau t^{2}D^{2}} forms a smooth section of the rescaled bundle.

This paper is organized as follows: in section 2 we review some basic facts of the theory of distributions on manifolds with coefficient in vector bundles and their Fourier transformations; in section 3 we summarize the construction of the rescaled bundle over tangent groupoid for closed spin manifold; in section 4 we recall the theory of fibered distributions developed in [LMV17] and, as in [EY17], use it to define a class of pseudodifferential operators in section 5; in section 6, we study the space of Taylor’s expansion of full symbols of pseudodifferential operators and show that this space has an explicitly computable multiplication formula which can be applied to heat equation and gives the asymptotic expansion of the heat kernel in section 7.

2 Distributions with coefficient in vector bundle

Let MM be a closed Riemannian manifold, EME\to M a vector bundle over MM and EE^{\ast} the dual vector bundle. The space of distributions with coefficient in EE denoted by 𝒟(M,E)\mathcal{D}^{\prime}(M,E) is defined to be the continuous dual of Cc(M,E)C_{c}^{\infty}(M,E^{\ast}). Notice that we have choose a Riemannian structure on MM so that the density is omitted in the discussion of distributions. Let {Ui}\{U_{i}\} be an open cover of MM that consists of finitely many open sets and EE is trivial over each member, denote by φi:E|UiUi×n\varphi_{i}:E|_{U_{i}}\to U_{i}\times\mathbb{R}^{n} the trivialization and ρi\rho_{i} the subordinate partition of unity. For any ui𝒟(M,E)u_{i}\in\mathcal{D}^{\prime}(M,E) that has supported within some UiU_{i}, we have ui𝒟(Ui)nu_{i}\in\mathcal{D}^{\prime}(U_{i})\otimes\mathbb{R}^{n}. In another word, uiu_{i} can be written as I=1nui,Iei,I\sum_{I=1}^{n}u_{i,I}\otimes e_{i,I} where ui,Iu_{i,I} are distributions on UiU_{i} and ei,Ie_{i,I} are basis of n\mathbb{R}^{n}.

For general u𝒟(M,E)u\in\mathcal{D}^{\prime}(M,E), we have the following decomposition:

(2.1) u=iρiu=iI=1nρiui,Iei,I.u=\sum_{i}\rho_{i}u=\sum_{i}\sum_{I=1}^{n}\rho_{i}u_{i,I}\otimes e_{i,I}.

Let u𝒟(M)u\in\mathcal{D}^{\prime}(M) be a distribution on manifold MM and φC(M,E)\varphi\in C^{\infty}(M,E) be a smooth section of EE, the product uφ𝒟(M,E)u\cdot\varphi\in\mathcal{D}^{\prime}(M,E) is defined so that

(2.2) uφ,s=u,φ,s\langle u\cdot\varphi,s\rangle=\langle u,\langle\varphi,s\rangle\rangle

for all sCc(M,E)s\in C^{\infty}_{c}(M,E^{\ast}). Here φ,s\langle\varphi,s\rangle denote the compactly supported function given by φ,s(m)=φ(m),s(m)E,E\langle\varphi,s\rangle(m)=\langle\varphi(m),s(m)\rangle_{E,E^{\ast}}. Moreover, it is straightforward to check that for any fC(M)f\in C^{\infty}(M), we have ufφ=ufφuf\cdot\varphi=u\cdot f\varphi. Under the light of (2.2), the equation (2.1) becomes u=i,Iρiui,Iei,Iu=\sum_{i,I}\rho_{i}u_{i,I}\cdot e_{i,I}. Therefore, we have the following proposition.

Proposition 2.1.

There is a finite set of sections sis_{i} of EME\to M such that any distribution u𝒟(M,E)u\in\mathcal{D}^{\prime}(M,E) can be written as

u=uisiu=\sum u_{i}\cdot s_{i}

where ui𝒟(M)u_{i}\in\mathcal{D}^{\prime}(M) are some scalar distributions on MM. ∎

Let π:TMM\pi:TM\to M be the tangent bundle of MM, the following result is parallel to proposition 2.1.

Proposition 2.2.

There is a finite set of sections sis_{i} of EME\to M such that any distribution u𝒟(TM,πE)u\in\mathcal{D}^{\prime}(TM,\pi^{\ast}E) can be written as

u=uiπsiu=\sum u_{i}\cdot\pi^{\ast}s_{i}

where ui𝒟(TM)u_{i}\in\mathcal{D}^{\prime}(TM) are some scalar distributions on TMTM and πsi\pi^{\ast}s_{i} are sections of πETM\pi^{\ast}E\to TM given by πsi(Xm)=si(m)\pi^{\ast}s_{i}(X_{m})=s_{i}(m) for all XmTmMX_{m}\in T_{m}M. Moreover, if uu is a tempered distribution (distribution with compact support respectively), then uiu_{i}’s can be chosen to be tempered distributions (distributions with compact support respectively) on the tangent bundle.

Proof.

If uu is a tempered distribution with support within coordinate chart UiU_{i}, then u𝒮(Ui×n)nu\in\mathcal{S}^{\prime}(U_{i}\times\mathbb{R}^{n})\otimes\mathbb{R}^{n}. A choice of partition of unity gives the same decomposition as (2.1) with coefficients ui,Iu_{i,I} scalar tempered distributions. The same argument works when uu is a distribution or a distribution with compact support. ∎

In this paper, we shall have occasion to consider the Fourier transformation of distributions with coefficient in vector bundles. Let u𝒮(TM,πE)u\in\mathcal{S}^{\prime}(TM,\pi^{\ast}E) be a tempered distribution with coefficient in EE, its Fourier transformation, which is denote by u^𝒮(TM,πE)\widehat{u}\in\mathcal{S}^{\prime}(T^{\ast}M,\pi^{\ast}E), is defined, according to [H0̈3, Chapter VII], to satisfy

u^,φ=u,φ^\langle\widehat{u},\varphi\rangle=\langle u,\widehat{\varphi}\rangle

where φ𝒮(TM,πE)\varphi\in\mathcal{S}(T^{\ast}M,\pi^{\ast}E^{\ast}) is any Schwartz section and φ^𝒮(TM,πE)\widehat{\varphi}\in\mathcal{S}(TM,\pi^{\ast}E^{\ast}) its Fourier transformation.

Proposition 2.3.

Under the light of Proposition 2.2, the Fourier transformation of tempered distributions u^\widehat{u} can be written as

u^=iu^iπsi.\widehat{u}=\sum_{i}\widehat{u}_{i}\cdot\pi^{\ast}s_{i}.

where u^i\widehat{u}_{i} is the Fourier transformation of the scalar tempered distribution uiu_{i}. Notice here we abuse the notation πsi\pi^{\ast}s_{i} to denote both the pullback of sis_{i} to the tangent bundle and the cotangent bundle.

Proof.

Let φ\varphi be any Schwartz section of πETM\pi^{\ast}E^{\ast}\to T^{\ast}M, then

u,φ^\displaystyle\langle u,\widehat{\varphi}\rangle =uiπsi,φ^\displaystyle=\langle\sum u_{i}\cdot\pi^{\ast}s_{i},\widehat{\varphi}\rangle
=iui,πsi,φ^.\displaystyle=\sum_{i}\langle u_{i},\langle\pi^{\ast}s_{i},\widehat{\varphi}\rangle\rangle.

According to the definition of πsi\pi^{\ast}s_{i} and the Fourier transformation, we have

πsi,φ^(Xm)=si(m),φ^(Xm)=(πsi,φ)(Xm)\langle\pi^{\ast}s_{i},\widehat{\varphi}\rangle(X_{m})=\langle s_{i}(m),\widehat{\varphi}(X_{m})\rangle=\mathcal{F}\left(\langle\pi^{\ast}s_{i},\varphi\rangle\right)(X_{m})

here ()\mathcal{F}(\cdot) denote the Fourier transformation of some Schwartz functions. The above equation continues

=iu^i,πsi,φ=iu^iπsi,φ=\sum_{i}\langle\widehat{u}_{i},\langle\pi^{\ast}s_{i},\varphi\rangle\rangle=\langle\sum_{i}\widehat{u}_{i}\cdot\pi^{\ast}s_{i},\varphi\rangle

which completes the proof. ∎

3 Rescaled bundle

Let UTMU\subset TM be an open neighborhood of the zero section where the exponential map is well defined and injective. Let 𝕌~\widetilde{\mathbb{U}} be the open neighborhood of TM×{0}TM×TM\times\{0\}\subset TM\times\mathbb{R} given by

𝕌~={(x,Y,t)TM×(x,tY)U}.\widetilde{\mathbb{U}}=\{(x,Y,t)\in TM\times\mathbb{R}\mid(x,-tY)\in U\}.

Then the following homeomorphism determines the smooth structure of the tangent groupoid near t=0t=0.

(3.1) TM×𝕌~\displaystyle TM\times\mathbb{R}\supset\widetilde{\mathbb{U}} Φ𝕌𝕋M\displaystyle\xrightarrow{\Phi}\mathbb{U}\subset\mathbb{T}M
(x,Y,t)\displaystyle(x,Y,t) (x,expx(tY),t)\displaystyle\mapsto(x,\exp_{x}(-tY),t)
(x,Y,0)\displaystyle(x,Y,0) (x,Y,0),\displaystyle\mapsto(x,Y,0),

where 𝕌𝕋M\mathbb{U}\subset\mathbb{T}M is the image of the above map.

Let MM be a closed spin manifold with spinor bundle SMS\to M. The rescaled bundle 𝕊𝕋M\mathbb{S}\to\mathbb{T}M is a vector bundle over the tangent groupoid whose restriction to M×M×M\times M\times\mathbb{R}^{\ast} is the pullback of SSS^{\ast}\boxtimes S and whose restriction to TMTM is the pullback of the bundle of exterior algebras. This information is summarized in the diagram (1.2). Each part in the diagram has its own smooth structure, to specify the smooth structure of the rescaled bundle it is enough to specify how does these two parts fit together.

Definition 3.1.

Denote by A(𝕋M)C(M×M)[t1,t]A(\mathbb{T}M)\subseteq C^{\infty}(M\times M)[t^{-1},t] the \mathbb{R}-algebra of those Laurent polynomials

(3.2) pfptp\sum_{p\in\mathbb{Z}}f_{p}t^{-p}

for which each coefficient fpf_{p} is a smooth, real-valued function on M×MM\times M that vanishes to order pp on the diagonal MM×MM\hookrightarrow M\times M.

Laurent polynomials (3.2) can be evaluated at any γ𝕋M\gamma\in\mathbb{T}M. The evaluation is given as follows: its value at γ=(x,y,λ)M×M×\gamma=(x,y,\lambda)\in M\times M\times\mathbb{R}^{\ast} is given by fp(x,y)λp\sum f_{p}(x,y)\lambda^{-p} and its value at (m,X,0)TM×{0}(m,X,0)\in TM\times\{0\} is given by1p!Xpfp\sum\frac{1}{p!}X^{p}f_{p}. Moreover, the character spectrum of the associative algebra AA is precisely the tangent groupoid 𝕋M\mathbb{T}M, namely for any point γ𝕋M\gamma\in\mathbb{T}M there is a maximal ideal IγA(𝕋M)I_{\gamma}\subset A(\mathbb{T}M) given by the collection of elements whose evaluation is zero at γ\gamma, and this viewpoint can be used to determines the manifold structure of 𝕋M\mathbb{T}M (see [HSSH18] or [HY19]).

One may take the algebra A(𝕋M)A(\mathbb{T}M) as the space of ”algebraic functions” on the tangent groupoid. In the following, we shall define an A(𝕋M)A(\mathbb{T}M)-module denoted by S(𝕋M)S(\mathbb{T}M) which can be taken as, with the previous analogy, the space of ”algebraic sections” of rescaled bundle.

Definition 3.2.

Denote by S(𝕋M)S(\mathbb{T}M) the complex vector space of Laurent polynomials

pσptp\sum_{p\in\mathbb{Z}}\sigma_{p}t^{-p}

where each σp\sigma_{p} is a smooth section of SSS^{\ast}\boxtimes S of scaling order at least pp.

We shall refer to [HY19, Definition 3.3.5] for a precise definition and detailed discussion for the notion of scaling order. However, we shall give an important example of sections with certain scaling order.

Example 3.3.

Let n=dim(M)n=\dim(M), VV be an open coordinate chart of MM over which the tangent bundle is trivial and e1,e2,,ene_{1},e_{2},\cdots,e_{n} is an orthonormal frame for TM|VTM|_{V}. For any multi-index I={i1,i2,,id}I=\{i_{1},i_{2},\cdots,i_{d}\} of length (I)=d\ell(I)=d, denote by eIe_{I} the Clifford multiplication

(3.3) ei1ei2eide_{i_{1}}e_{i_{2}}\cdots e_{i_{d}}

which is a local section of the bundle of Clifford algebras Cliff(TM)M\operatorname{Cliff}(TM)\to M. Taken as the restriction of SSM×MS^{\ast}\boxtimes S\to M\times M to the diagonal, we can use parallel translation in the second variable to extend the local section eIe_{I} to a local section of SSS^{\ast}\boxtimes S. We shall use the same notation eIe_{I} to denote the extension, and it has scaling order (I)-\ell(I). In fact, any local section σ\sigma of SSS^{\ast}\boxtimes S has the form

σ=IfIeI\sigma=\sum_{I}f_{I}e_{I}

where fIf_{I} are smooth functions on M×MM\times M. If the smooth function fIf_{I} vanishes to order p+(I)p+\ell(I) for all II then σ\sigma has scaling order pp.

It is a fact that S(𝕋M)S(\mathbb{T}M) is a module over A(𝕋M)A(\mathbb{T}M) by ordinary multiplication of Laurent polynomials. The fiber of S(𝕋M)S(\mathbb{T}M) over γ\gamma is defined to be

S(𝕋M)|γ=S(𝕋M)/IγS(𝕋M).S(\mathbb{T}M)|_{\gamma}=S(\mathbb{T}M)\big{/}I_{\gamma}\cdot S(\mathbb{T}M).

There is an isomorphism between S(𝕋M)|(x,y,t)S(\mathbb{T}M)|_{(x,y,t)} and SxSyS^{\ast}_{x}\otimes S_{y} which is induced by the map

ε(x,y,λ):sptpsp(x,y)λp\varepsilon_{(x,y,\lambda)}:\sum s_{p}t^{-p}\mapsto\sum s_{p}(x,y)\lambda^{-p}

for t0t\neq 0. For t=0t=0, there is a similar isomorphism between S(𝕋M)|(x,Y,0)S(\mathbb{T}M)|_{(x,Y,0)} and TxM\wedge T^{\ast}_{x}M which is slightly more complicated and we refer the reader to [HY19, Proposition 3.4.9]. According to these two isomorphisms, there is a map

S(𝕋M)γ𝕋MS(𝕋M)|γS(\mathbb{T}M)\to\prod_{\gamma\in\mathbb{T}M}S(\mathbb{T}M)|_{\gamma}

which sends σS(𝕋M)\sigma\in S(\mathbb{T}M) to its image σ^\widehat{\sigma} under the corresponding isomorphism.

Definition 3.4.

Denote by 𝑺𝕋M\boldsymbol{{S}}_{\mathbb{T}M} the sheaf on 𝕋M\mathbb{T}M consisting of sections

𝕋Mγτ(γ)S(𝕋M)|γ\mathbb{T}M\ni\gamma\longmapsto\tau(\gamma)\in S(\mathbb{T}M)|_{\gamma}

that are locally of the form

τ(γ)=j=1Nfj(γ)σ^j(γ)\tau(\gamma)=\sum_{j=1}^{N}f_{j}(\gamma)\cdot\widehat{\sigma}_{j}(\gamma)

for some NN\in\mathbb{N}, where f1,,fNf_{1},\dots,f_{N} are smooth functions on 𝕋M\mathbb{T}M and σ1,,σN\sigma_{1},\dots,\sigma_{N} belong to S(𝕋M)S(\mathbb{T}M).

It can be shown that this is a locally free sheaf and is the space of smooth sections of the rescaled bundle 𝕊𝕋M\mathbb{S}\to\mathbb{T}M. According to the construction, for any element σ\sigma of S(𝕋M)S(\mathbb{T}M), the assignment σ^\widehat{\sigma} which sends γ𝕋M\gamma\in\mathbb{T}M to σ^(γ)\widehat{\sigma}(\gamma) is a smooth section of the rescaled bundle. In fact, thanks to Example 3.3, the assignments eIt(I)^\widehat{e_{I}t^{\ell(I)}}, as II ranges over all multi-index I{1,2,,n}I\subset\{1,2,\cdots,n\}, form a local frame of the rescaled bundle. The following Proposition clarify the smooth structure of the rescaled bundle over 𝕌\mathbb{U}.

Proposition 3.5.

The pullback of the rescaled bundle along Φ:𝕌~𝕋M\Phi:\widetilde{\mathbb{U}}\to\mathbb{T}M is isomorphic to the pullback of the bundle of exterior algebras along ρ:TM×𝕌~M\rho:TM\times\mathbb{R}\supset\widetilde{\mathbb{U}}\to M.

Proof.

Let VV be a finite dimensional vector space with an inner product and an orthonormal basis e1,e2,,ene_{1},e_{2},\cdots,e_{n}. Denote by qt:VCliff(V)q_{t}:\wedge^{\ast}V\to\operatorname{Cliff}(V) the quantization map which is given by

qt(ei1ei2eik)=tkei1ei2eikq_{t}(e_{i_{1}}\wedge e_{i_{2}}\wedge\cdots\wedge e_{i_{k}})=t^{k}e_{i_{1}}e_{i_{2}}\cdots e_{i_{k}}

for all t0t\neq 0. Denote by σt:Cliff(V)V\sigma_{t}:\operatorname{Cliff}(V)\to\wedge^{\ast}V the symbol map which is the inverse of qtq_{t}. Let S:ρTM𝕊|𝕌S:\rho^{\ast}\wedge T^{\ast}M\to\mathbb{S}|_{\mathbb{U}} be the map given by

(ρTM)(x,Y,t)ωτ2(x,expx(tY))qtω𝕊(x,expx(tY),t)\left(\rho^{\ast}\wedge T^{\ast}M\right)_{(x,Y,t)}\ni\omega\mapsto\tau_{2}(x,\exp_{x}(-tY))q_{t}\omega\in\mathbb{S}_{(x,\exp_{x}(-tY),t)}

when t0t\neq 0, and

(ρTM)(x,Y,0)ωω𝕊(x,Y,0).\left(\rho^{\ast}\wedge T^{\ast}M\right)_{(x,Y,0)}\ni\omega\mapsto\omega\in\mathbb{S}_{(x,Y,0)}.

here we thought qtωq_{t}\omega as an element in 𝕊(x,x,t)\mathbb{S}_{(x,x,t)} and τ2(x,expx(tY))\tau_{2}(x,\exp_{x}(-tY)) is the parallel translation of the bundle SSS\boxtimes S^{\ast} in the second variable from xx to expx(tY)\exp_{x}(-tY). We can also define the inverse map T:𝕊|𝕌ρTMT:\mathbb{S}|_{\mathbb{U}}\to\rho^{\ast}\wedge T^{\ast}M which sends

𝕊(x,expx(tY),t)ωσt(τ2(expx(tY),x)ω)(ρTM)(x,Y,t)\mathbb{S}_{(x,\exp_{x}(-tY),t)}\ni\omega\mapsto\sigma_{t}\left(\tau_{2}(\exp_{x}(-tY),x)\omega\right)\in\left(\rho^{\ast}\wedge T^{\ast}M\right)_{(x,Y,t)}

for t0t\neq 0, and

𝕊(x,Y,0)ωω(ρTM)(x,Y,0),\mathbb{S}_{(x,Y,0)}\ni\omega\mapsto\omega\in\left(\rho^{\ast}\wedge T^{\ast}M\right)_{(x,Y,0)},

here τ2(expx(tY),x)\tau_{2}(\exp_{x}(-tY),x) is the parallel translation of SSS\boxtimes S^{\ast} in the second variable from expx(tY)\exp_{x}(-tY) to xx. It is easy to check that these two maps are mutually inverse.

It remains to show that TT and SS are smooth. Indeed, let VMV\subset M be an open subset over which the tangent bundle TMTM is trivial and the trivialization is given by an orthonormal frame e1,e2,,ene_{1},e_{2},\cdots,e_{n}. Let I={i1,i2,,i(I)}{1,2,,n}I=\{i_{1},i_{2},\cdots,i_{\ell(I)}\}\subset\{1,2,\cdots,n\} be the multi-index, denote by Ie\wedge^{I}e the form ei1ei2ei(I)e_{i_{1}}\wedge e_{i_{2}}\wedge\cdots\wedge e_{i_{\ell(I)}} and eIe_{I} the Clifford multiplication ei1ei2ei(I)e_{i_{1}}e_{i_{2}}\cdots e_{i_{\ell(I)}}. Then the bundle (ρTM)|(TV×)𝕌~\left(\rho^{\ast}\wedge T^{\ast}M\right)|_{(TV\times\mathbb{R})\cap\widetilde{\mathbb{U}}} is also trivial and the constant sections {Ie}I\{\wedge^{I}e\}_{I} form a local orthonormal frame. According to the definition, SS sends the constant section Ie\wedge^{I}e to the smooth section eIt(I)^\widehat{e_{I}t^{\ell(I)}} of 𝕊\mathbb{S} which implies the smoothness of SS. The smoothness of TT can be verified in the same way. ∎

The rescaled bundle also has a multiplicative structure which we shall describe now. Let GG be a Lie groupoid, G(2)={(γ,η)G×Gs(γ)=r(η)}G^{(2)}=\{(\gamma,\eta)\in G\times G\mid s(\gamma)=r(\eta)\} be the set of composable pairs. Denote by p1,p2:G(2)Gp_{1},p_{2}:G^{(2)}\to G the two coordinate projections and m:G(2)Gm:G^{(2)}\to G the multiplication map.

Definition 3.6.

Let EGE\to G be a vector bundle. We shall say that EE has a multiplicative structure if there is a bundle map (which we shall also denote by mm)

(3.4) p1Ep2E\textstyle{p_{1}^{\ast}E\otimes p_{2}^{\ast}E\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}m\scriptstyle{\quad m}E\textstyle{E\ignorespaces\ignorespaces\ignorespaces\ignorespaces}G(2)\textstyle{G^{(2)}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}m\scriptstyle{m}G\textstyle{G}

that covers the multiplication map of the groupoid G(2)GG^{(2)}\to G. The bundle p1Ep2Ep_{1}^{\ast}E\otimes p_{2}^{\ast}E is sometimes denoted by EEE\boxtimes E.

The multiplicative structure of the rescaled bundle m:p1𝕊p2𝕊𝕊m:p_{1}^{\ast}\mathbb{S}\otimes p_{2}^{\ast}\mathbb{S}\to\mathbb{S} is given as follows:

  • the restriction of mm away from t=0t=0 is the multiplication 𝕊(x,y,t)𝕊(y,z,t)𝕊(x,z,t)\mathbb{S}_{(x,y,t)}\otimes\mathbb{S}_{(y,z,t)}\to\mathbb{S}_{(x,z,t)} given by contracting the middle two variables;

  • the restriction of mm to t=0t=0 is the multiplication 𝕊(x,Y,0)𝕊(x,Z,0)𝕊(x,Y+Z,0)\mathbb{S}_{(x,Y,0)}\otimes\mathbb{S}_{(x,Z,0)}\to\mathbb{S}_{(x,Y+Z,0)} given by sending αβTxMTxM\alpha\otimes\beta\in\wedge T^{\ast}_{x}M\otimes\wedge T^{\ast}_{x}M to αβexp(12κ(Y,Z))\alpha\wedge\beta\wedge\exp(-\frac{1}{2}\kappa(Y,Z)) where κ(Y,Z)\kappa(Y,Z) is the differential two form given by the symbol of curvature of the spinor bundle.

(See [HY19, Section 4])

4 Fibered distributions

Most of the statements in this section are valid for general vector bundle over Lie groupoid that has a multiplicative structure. At the end of this section, we shall specialize to the case when G=𝕋MG=\mathbb{T}M is the tangent groupoid and E=𝕊E=\mathbb{S} is the rescaled bundle.

Definition 4.1.

Let s:MBs:M\to B be a submersion, a ss-fibered distribution on MM with coefficient in a vector bundle EME\to M is a continuous C(B)C^{\infty}(B)-module map

u:Cc(M,E)C(B)u:C^{\infty}_{c}(M,E^{\ast})\to C^{\infty}(B)

where the C(B)C^{\infty}(B)-module structure on Cc(M,E)C^{\infty}_{c}(M,E^{\ast}) is given by the submersion ss. The set of all ss-fibered distribution is denoted by 𝒟s(M,E)\mathcal{D}^{\prime}_{s}(M,E). We can also define ss-fibered compactly supported distribution on MM with coefficient in EE which is given by a continuous C(B)C^{\infty}(B)-module map

(4.1) u:C(M,E)C(B).u:C^{\infty}(M,E^{\ast})\to C^{\infty}(B).

The set of all such ss-fibered distributions will be denoted by s(M,E)\mathcal{E}^{\prime}_{s}(M,E). If MM is a vector bundle over BB and ss is the bundle projection, one can talk about the ss-fibered tempered distribution with coefficient in EME\to M given by a continuous C(B)C^{\infty}(B)-module map

u:𝒮(M,E)C(B).u:\mathcal{S}(M,E^{\ast})\to C^{\infty}(B).

The set of all ss-fibered tempered distributions will be denoted by 𝒮s(M,E)\mathcal{S}_{s}^{\prime}(M,E).

In the following discussion, we shall concern mainly with the case when MM is a Lie groupoid GG, BB is the unit space G(0)G^{(0)} and ss is the range map r:GG(0)r:G\to G^{(0)}. We shall also have occasion to consider the case when M=G(2)M=G^{(2)} with vector bundle EEG(2)E\boxtimes E\to G^{(2)}, B=G(0)B=G^{(0)} and s=rms=r\circ m the composition of the multiplication map with the range map.

Remark 4.2.

This definition of fibered compactly supported distribution is used in [EY17] which is slightly different from the one used in [LMV17] where the authors defined the compactly supported fibered distribution by the continuous C(B)C^{\infty}(B)-module maps

(4.2) u:C(M)Cc(B).u:C^{\infty}(M)\to C_{c}^{\infty}(B).

The continuity of u:C(M)Cc(B)u:C^{\infty}(M)\to C^{\infty}_{c}(B) implies that there is a constant CC, a compact subset KGK\subset G and a finite set of seminorms pi(f)=supγK|αf(γ)|p_{i}(f)=\sup_{\gamma\in K}\left|\partial^{\alpha}f(\gamma)\right| on C(M)C^{\infty}(M) such that

supxB|u(f)(x)|Cpi(f)\sup_{x\in B}\left|u(f)(x)\right|\leq C\cdot\sum p_{i}(f)

for all fC(M)f\in C^{\infty}(M). This implies that uu is supported within KK. The definition we use here, on the other hand, does not require uu to have overall compact support.

Remark 4.3.

If u𝒟s(M,E)u\in\mathcal{D}^{\prime}_{s}(M,E) (or s(M,E)\mathcal{E}^{\prime}_{s}(M,E) respectively), let fCc(M,E)f\in C^{\infty}_{c}(M,E^{\ast}) (or C(M,E)C^{\infty}(M,E^{\ast}) respectively), we shall write

u,f(x)=ux,f|s1(x)\langle u,f\rangle(x)=\langle u^{x},f|_{s^{-1}(x)}\rangle

where xBx\in B and uxu^{x} is the induced distribution on the fiber s1(x)s^{-1}(x). So, we may take the ss-fibered distribution uu as a family of distributions. It is easy to see that uxu^{x} are all compactly supported when us(M,E)u\in\mathcal{E}^{\prime}_{s}(M,E).

Now, let EGE\to G be a vector bundle with multiplicative structure, we shall describe the algebra structure on the set of distributions r(G,E)\mathcal{E}^{\prime}_{r}(G,E). The multiplicative structure (3.4) of EE induces a map m:C(G,E)C(G(2),EE)m^{\ast}:C^{\infty}(G,E^{\ast})\to C^{\infty}(G^{(2)},E^{\ast}\boxtimes E^{\ast}) which is given by mf(γ1,γ2)=mγ1,γ2f(γ1γ2)m^{\ast}f(\gamma_{1},\gamma_{2})=m_{\gamma_{1},\gamma_{2}}^{\ast}f(\gamma_{1}\gamma_{2}), here ff is any smooth section of EGE^{\ast}\to G and mγ1,γ2:Eγ1γ2Eγ1Eγ2m_{\gamma_{1},\gamma_{2}}^{\ast}:E^{\ast}_{\gamma_{1}\gamma_{2}}\to E^{\ast}_{\gamma_{1}}\otimes E^{\ast}_{\gamma_{2}} is the dual of the multiplication map m:Eγ1Eγ2Eγ1γ2m:E_{\gamma_{1}}\otimes E_{\gamma_{2}}\to E_{\gamma_{1}\gamma_{2}}. At the level of fibered distributions, we can define m:rm(G(2),EE)r(G,E)m_{\ast}:\mathcal{E}^{\prime}_{r\circ m}(G^{(2)},E\boxtimes E)\to\mathcal{E}^{\prime}_{r}(G,E) by mu(f)=u(mf),m_{\ast}u(f)=u(m^{\ast}f), for any fC(G,E)f\in C^{\infty}(G,E^{\ast}) and any urm(G(2),EE)u\in\mathcal{E}^{\prime}_{r\circ m}(G^{(2)},E\boxtimes E). Next, consider the commutative diagram

G(2)\textstyle{G^{(2)}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}p2\scriptstyle{p_{2}}p1\scriptstyle{p_{1}}G\textstyle{G\ignorespaces\ignorespaces\ignorespaces\ignorespaces}s\scriptstyle{s}G\textstyle{G\ignorespaces\ignorespaces\ignorespaces\ignorespaces}r\scriptstyle{r}G(0).\textstyle{G^{(0)}.}

Notice that the fibers of p1:G(2)Gp_{1}:G^{(2)}\to G correspond to the range fibers of GG. For any ur(G,E)u\in\mathcal{E}^{\prime}_{r}(G,E), its pullback p1up_{1}^{\ast}u by the first projection p1p_{1} is defined to be the continuous C(G)C^{\infty}(G)-module map C(G(2),EE)C(G,E)C^{\infty}(G^{(2)},E^{\ast}\boxtimes E^{\ast})\to C^{\infty}(G,E^{\ast}) which is given by

p1u,g(γ)=us(γ),g|(γ,Gs(γ))\langle p_{1}^{\ast}u,g\rangle(\gamma)=\langle u^{s(\gamma)},g|_{(\gamma,G^{s(\gamma)})}\rangle

here the C(G)C^{\infty}(G)-module structure on C(G(2),EE)C^{\infty}(G^{(2)},E^{\ast}\boxtimes E^{\ast}) is given by p1p_{1}, we take (γ,Gs(γ))(\gamma,G^{s(\gamma)}) as a submanifold of G(2)G^{(2)} and us(γ):C(Gs(γ),E)u^{s(\gamma)}:C^{\infty}(G^{s(\gamma)},E^{\ast})\to\mathbb{C} is the distribution induced by uu.

Remark 4.4.

In fact, if u𝒟r(G,E)u\in\mathcal{D}^{\prime}_{r}(G,E), the same construction p1up_{1}^{\ast}u is a C(G)C^{\infty}(G)-module map that sends Cc(G(2),EE)C^{\infty}_{c}(G^{(2)},E^{\ast}\boxtimes E^{\ast}) to Cc(G,E)C^{\infty}_{c}(G,E^{\ast}).

Let u,vr(G,E)u,v\in\mathcal{E}^{\prime}_{r}(G,E), then the convolution multiplication uvu\ast v is defined to be m(upr2v)r(G,E)m_{\ast}(u\circ\operatorname{pr_{2}}^{\ast}v)\in\mathcal{E}^{\prime}_{r}(G,E). Explicitly the convolution is given by

(4.3) uv,φ(x)=ux(γ),vs(γ)(η),mφ(γ,η)\langle u\ast v,\varphi\rangle(x)=\left\langle u^{x}(\gamma),\left\langle v^{s(\gamma)}(\eta),m^{\ast}\varphi(\gamma,\eta)\right\rangle\right\rangle

for φC(G,E)\varphi\in C^{\infty}(G,E^{\ast}). In fact, slightly more is true.

Corollary 4.5.

If u,v𝒟r(G,E)u,v\in\mathcal{D}^{\prime}_{r}(G,E) and at least one of them belong to r(G,E)\mathcal{E}^{\prime}_{r}(G,E) and is properly supported, then uvu\ast v given by the formula (4.3) is still well-defined and belongs to 𝒟r(G,E)\mathcal{D}^{\prime}_{r}(G,E).

Proof.

If vr(G,E)v\in\mathcal{E}_{r}^{\prime}(G,E) and is properly supported, u𝒟r(G,E)u\in\mathcal{D}_{r}^{\prime}(G,E), according to the Remark 4.4, the composition up1vu\circ p_{1}^{\ast}v belongs to 𝒟rm(G(2),EE)\mathcal{D}^{\prime}_{r\circ m}(G^{(2)},E\boxtimes E). As the multiplication map m:G(2)Gm:G^{(2)}\to G is not necessarily proper, there is no guarantee that the map m:rm(G(2),EE)r(G,E)m_{\ast}:\mathcal{E}^{\prime}_{r\circ m}(G^{(2)},E\boxtimes E)\to\mathcal{E}^{\prime}_{r}(G,E) can be extended to 𝒟rm(G(2),EE)𝒟r(G,E)\mathcal{D}^{\prime}_{r\circ m}(G^{(2)},E\boxtimes E)\to\mathcal{D}^{\prime}_{r}(G,E). To proceed, we need to use the property of vv in a more serious way.

According to the assumption on the support of vv, there is a properly support function φ\varphi with φv=v\varphi\cdot v=v and up1v=p2φ(up1v)u\circ p_{1}^{\ast}v=p_{2}^{\ast}\varphi\cdot(u\circ p_{1}^{\ast}v). Consider the map C(G,E)C(G(2),EE)C^{\infty}(G,E^{\ast})\to C^{\infty}(G^{(2)},E^{\ast}\boxtimes E^{\ast}) which sends fC(G,E)f\in C^{\infty}(G,E^{\ast}) to p2φmfp_{2}^{\ast}\varphi\cdot m^{\ast}f. If ff is compactly supported, then the support of p2φmfp_{2}^{\ast}\varphi\cdot m^{\ast}f is p21(suppφ)m1(suppf)p_{2}^{-1}(\operatorname{supp}{\varphi})\cap m^{-1}(\operatorname{supp}{f}). Notice that sm=sp2s\circ m=s\circ p_{2}, and φ\varphi is properly supported, it follows that p2φmfp^{\ast}_{2}\varphi\cdot m^{\ast}f is also compactly supported.

Thus the above map takes the subspace Cc(G,E)C^{\infty}_{c}(G,E^{\ast}) into the subspace Cc(G(2),EE)C_{c}^{\infty}(G^{(2)},E^{\ast}\boxtimes E^{\ast}). Under this light, we have the equation

m(p2φ(up1v))f=(up1v)(p2φmf)m_{\ast}\left(p_{2}^{\ast}\varphi\cdot(u\circ p_{1}^{\ast}v)\right)f=\left(u\circ p_{1}^{\ast}v\right)(p_{2}^{\ast}\varphi\cdot m^{\ast}f)

which proves m(up1v)𝒟r(G,E)m_{\ast}(u\circ p_{1}^{\ast}v)\in\mathcal{D}^{\prime}_{r}(G,E).

The other case, when ur(G,E)u\in\mathcal{E}_{r}^{\prime}(G,E) and is properly supported, v𝒟r(G,E)v\in\mathcal{D}_{r}^{\prime}(G,E), can be proved in a similar way, except this time, there is a properly supported function with u=φuu=\varphi\cdot u and up1v=p1φ(up1v)u\circ p^{\ast}_{1}v=p_{1}^{\ast}\varphi\cdot(u\circ p_{1}^{\ast}v). The fact that rm=rp1r\circ m=r\circ p_{1} implies that the map C(G,E)C(G(2),EE)C^{\infty}(G,E^{\ast})\to C^{\infty}(G^{(2)},E^{\ast}\boxtimes E^{\ast}) which sends fC(G,E)f\in C^{\infty}(G,E^{\ast}) to p1φmfC(G(2),EE)p_{1}^{\ast}\varphi\cdot m^{\ast}f\in C^{\infty}(G^{(2)},E^{\ast}\boxtimes E^{\ast}) take the subspace of compactly supported sections to compactly supported sections. This completes the proof. ∎

Now set GG to be the tangent groupoid and EE to be the rescaled bundle.

Proposition 4.6.

Let u,vr(G,E)u,v\in\mathcal{E}^{\prime}_{r}(G,E), write u0u_{0} and v0v_{0} for the restrictions u|t=0u|_{t=0} and v|t=0v|_{t=0}, let u^0,v^0\widehat{u}_{0},\widehat{v}_{0} be their Fourier transformations, then the Fourier transformation of u0v0u_{0}\ast v_{0} admits the following expansion

(4.4) exp(12κ(ξ,η))u^0(ξ)v^0(η)|ξ=η\exp\left(-\frac{1}{2}\kappa(\partial_{\xi},\partial_{\eta})\right)\widehat{u}_{0}(\xi)\wedge\widehat{v}_{0}(\eta)|_{\xi=\eta}

which coincide with the formula in [Get83, Theorem 2.7]. In the following discussion, we shall write u^0#0v^0\widehat{u}_{0}\#_{0}\widehat{v}_{0} for (4.4).

Proof.

Let π:TMM\pi:TM\to M be the bundle projection, since u0,v0π(TM,πTM)u_{0},v_{0}\in\mathcal{E}_{\pi}^{\prime}(TM,\pi^{\ast}\wedge T^{\ast}M) , according to Proposition 2.2, we may assume that u0u_{0} and v0v_{0} have the following forms

u0=iuiπsiu_{0}=\sum_{i}u_{i}\cdot\pi^{\ast}s_{i}

and

v0=jvjπsjv_{0}=\sum_{j}v_{j}\cdot\pi^{\ast}s_{j}

where sis_{i} are some sections of the bundle of exterior algebras over MM. Then for any Schwartz section φ\varphi of πTMTM\pi^{\ast}\wedge T^{\ast}M\to TM we have

u0v0,φ^(x)\displaystyle\langle u_{0}\ast v_{0},\widehat{\varphi}\rangle(x) =u0x(X),v0x(Y),mX,Yφ^(X+Y)\displaystyle=\left\langle u_{0}^{x}(X),\left\langle v_{0}^{x}(Y),m^{\ast}_{X,Y}\widehat{\varphi}(X+Y)\right\rangle\right\rangle
=i,juix(X)πsi,vx(Y)πsj,mX,Yφ^(X+Y)\displaystyle=\sum_{i,j}\left\langle u_{i}^{x}(X)\cdot\pi^{\ast}s_{i},\left\langle v^{x}(Y)\cdot\pi^{\ast}s_{j},m_{X,Y}^{\ast}\widehat{\varphi}(X+Y)\right\rangle\right\rangle
=i,juix(X)vjx(Y),πsiπsj,mX,Yφ^(X+Y)\displaystyle=\sum_{i,j}\langle u^{x}_{i}(X)\otimes v^{x}_{j}(Y),\langle\pi^{\ast}s_{i}\otimes\pi^{\ast}s_{j},m^{\ast}_{X,Y}\widehat{\varphi}(X+Y)\rangle\rangle
=i,jexp(12κ(X,Y))uix(X)vjx(Y)πsiπsj,φ^(X+Y).\displaystyle=\sum_{i,j}\left\langle\exp\left(-\frac{1}{2}\kappa(X,Y)\right)u^{x}_{i}(X)\otimes v^{x}_{j}(Y)\pi^{\ast}s_{i}\wedge\pi^{\ast}s_{j},\widehat{\varphi}(X+Y)\right\rangle.

where from the third line to the fourth line, we use the definition of mX,Ym_{X,Y}^{\ast}. The function φ^(X+Y)\widehat{\varphi}(X+Y) when taking as a function with two variables F^(X,Y)\widehat{F}(X,Y), its inverse Fourier transformation on both variables is given by

F(ξ,η)\displaystyle F(\xi,\eta) =F^(X,Y)eiXξYη𝑑X𝑑Y\displaystyle=\int\widehat{F}(X,Y)e^{-iX\cdot\xi-Y\cdot\eta}dXdY
=φ^(X+Y)eiXξYη𝑑X𝑑Y\displaystyle=\int\widehat{\varphi}(X+Y)e^{-iX\cdot\xi-Y\cdot\eta}dXdY
=φ^(X)eiXξ𝑑XeiY(ηξ)𝑑Y\displaystyle=\int\widehat{\varphi}(X)e^{-iX\cdot\xi}dX\int e^{-iY\cdot(\eta-\xi)}dY
=φ(ξ)δξ=η\displaystyle=\varphi(\xi)\cdot\delta_{\xi=\eta}

According to the definition of Fourier transformation of compactly supported distributions, we conclude that the Fourier transformation of u0v0u_{0}\ast v_{0} is given by first take the Fourier transformation of

exp(12κ(X,Y))i,jui(X)vj(Y)πsiπsj.\exp(-\frac{1}{2}\kappa(X,Y))\sum_{i,j}u_{i}(X)\otimes v_{j}(Y)\cdot\pi^{\ast}s_{i}\wedge\pi^{\ast}s_{j}.

on both variables and then set ξ=η\xi=\eta. The formula is expressed precisely as in (4.4). ∎

Analogous to Corollary 4.5, the same proof results in the following Corollary.

Corollary 4.7.

Let u,v𝒟r(G,E)u,v\in\mathcal{D}^{\prime}_{r}(G,E), at least one of them belong to r(G,E)\mathcal{E}^{\prime}_{r}(G,E) with proper support and u0,v0u_{0},v_{0} both belong to 𝒮(TM,πTM)\mathcal{S}^{\prime}(TM,\pi^{\ast}\wedge T^{\ast}M), then the Fourier transformation of u0v0u_{0}\ast v_{0} is computed by the same formula (4.4).

5 Pseudodifferential operators on rescaled bundle

The following pseudo-local property of elements in Ψm(𝕋M,𝕊)\Psi^{m}(\mathbb{T}M,\mathbb{S}) is a consequence of the properness of their support and the essentially homogeneous condition (1.5).

Proposition 5.1.

[EY17, Proposition 22] Let Ψm(𝕋M,𝕊)\mathbb{P}\in\Psi^{m}(\mathbb{T}M,\mathbb{S}) for some mm\in\mathbb{R}, then \mathbb{P} is smooth on 𝕋M\𝕋M(0)\mathbb{T}M\backslash\mathbb{T}M^{(0)}. ∎

As a consequence, modulo the space of smooth sections, elements in Ψm(𝕋M,𝕊)\Psi^{m}(\mathbb{T}M,\mathbb{S}) may be assumed to have support within 𝕌\mathbb{U}.

Definition 5.2.

A smooth function φC(𝕋M)\varphi\in C^{\infty}(\mathbb{T}M) is called a cut-off function if it is identically one on some neighborhood of 𝕋M|t=0𝕋M(0)\mathbb{T}M|_{t=0}\cup\mathbb{T}M^{(0)} and supported within 𝕌\mathbb{U}.

Proposition 5.3.

[EY17, Lemma 27] There is a cut-off function φC(𝕋M)\varphi\in C^{\infty}(\mathbb{T}M) such that for any Ψm(𝕋M;𝕊),=φΨm(𝕋M;𝕊)\mathbb{P}\in\Psi^{m}(\mathbb{T}M;\mathbb{S}),\mathbb{P}^{\prime}=\varphi\cdot\mathbb{P}\in\Psi^{m}(\mathbb{T}M;\mathbb{S}) is supported within 𝕌\mathbb{U} and satisfies Cp(𝕋M;𝕊)\mathbb{P}-\mathbb{P}^{\prime}\in C^{\infty}_{p}(\mathbb{T}M;\mathbb{S}). ∎

The difference \mathbb{P}-\mathbb{P}^{\prime} is not only smooth properly supported, but also vanishes on some neighborhood of 𝕋M|t=0\mathbb{T}M|_{t=0}. In another word, the difference \mathbb{P}-\mathbb{P}^{\prime} belongs to the set

(5.1) A={Cp(𝕋M,𝕊) vanishes to infinite order at t=0}.A=\{\mathbb{P}\in C^{\infty}_{p}(\mathbb{T}M,\mathbb{S})\mid\mathbb{P}\text{ vanishes to infinite order at }t=0\}.

In the following discussion, rather than plain distributions in Ψm(𝕋M,𝕊)\Psi^{m}(\mathbb{T}M,\mathbb{S}), we shall focus on the space of equivalent classes Ψm(𝕋M,𝕊)/A\Psi^{m}(\mathbb{T}M,\mathbb{S})/A.

Proposition 5.4.

[EY17, Definition 41 and Proposition 42] Any equivalent class in the quotient space Ψm(𝕋M,𝕊)/A\Psi^{m}(\mathbb{T}M,\mathbb{S})/A admits a representative that is homogeneous on the nose outside [1,1][-1,1] and supported within 𝕌\mathbb{U}. In another word, there is a representative Ψm(𝕋M,𝕊)\mathbb{P}\in\Psi^{m}(\mathbb{T}M,\mathbb{S}) that is supported within 𝕌\mathbb{U} and satisfy t=tm1\mathbb{P}_{t}=t^{m}\mathbb{P}_{1} and t=tm1\mathbb{P}_{-t}=t^{m}\mathbb{P}_{-1} for all t>1t>1. ∎

In the following discussion, we shall always choose representatives that are homogeneous on the nose outside [1,1][-1,1] and have support within 𝕌\mathbb{U}. Combine with the Proposition 3.5, these choices enable us to do more explicit local calculation near 𝕌\mathbb{U}.

The α\alpha-action given below the equation (1.5) on the tangent groupoid is transformed, under the homeomorphism Φ\Phi defined in (3.1), into an action α~\widetilde{\alpha} on TM×TM\times\mathbb{R} which is given by

Φ1αλ(Φ(x,Y,t))=α~λ(x,Y,t).\Phi^{-1}\alpha_{\lambda}\left(\Phi(x,Y,t)\right)=\widetilde{\alpha}_{\lambda}(x,Y,t).

In another word, α~λ\widetilde{\alpha}_{\lambda} maps (x,Y,t)(x,Y,t) to (x,λ1Y,λt)(x,\lambda^{-1}Y,\lambda t).

We shall still use eIt(I)^\widehat{e_{I}t^{\ell(I)}} to denote the spanning sections of the pullback Φ𝕊𝕌~\Phi^{\ast}\mathbb{S}\to\widetilde{\mathbb{U}} whose value at (x,Y,t)(x,Y,t) is eI(x,expx(tY))t(I)e_{I}(x,\exp_{x}(-tY))t^{\ell(I)} and whose value at (x,Y,0)(x,Y,0) is Ie\wedge^{I}e. It can be checked that

(5.2) T(α~λ,eIt(I)^(x,Y,t))=λ(I)T(eIt(I)^(x,Y,t)).T\left(\widetilde{\alpha}_{\lambda,\ast}\widehat{e_{I}t^{\ell(I)}}(x,Y,t)\right)=\lambda^{\ell(I)}T\left(\widehat{e_{I}t^{\ell(I)}}(x,Y,t)\right).

Under the light of Proposition 3.5, \mathbb{P} is assumed to have support within 𝕌\mathbb{U}, the pullback Φ\Phi^{\ast}\mathbb{P} belongs to r(TM×;Φ𝕊)\mathcal{E}^{\prime}_{r}(TM\times\mathbb{R};\Phi^{\ast}\mathbb{S}) and can be written as

Φ=IuIeIt(I)^\Phi^{\ast}\mathbb{P}=\sum_{I}u_{I}\cdot\widehat{e_{I}t^{\ell(I)}}

where uIu_{I}’s are rr-fibered distributions on TM×TM\times\mathbb{R} that has support within 𝕌~\widetilde{\mathbb{U}}. We have T(Φ)r(TM×;ρTM)T(\Phi^{\ast}\mathbb{P})\in\mathcal{E}^{\prime}_{r}(TM\times\mathbb{R};\rho^{\ast}\wedge T^{\ast}M), thus its Fourier transformation

^=T(Φ)^=Iu^IT(eIt(I)^)\widehat{\mathbb{P}}=\widehat{T(\Phi^{\ast}\mathbb{P})}=\sum_{I}\widehat{u}_{I}\cdot T\left(\widehat{e_{I}t^{\ell(I)}}\right)

belongs to C(TM×;ρTM)C^{\infty}(T^{\ast}M\times\mathbb{R};\rho^{\ast}\wedge T^{\ast}M). The essentially homogeneous condition (1.5) now becomes

(5.3) βλ^λm^𝒮ρ(TM×;ρTM)\beta^{\ast}_{\lambda}\widehat{\mathbb{P}}-\lambda^{m}\widehat{\mathbb{P}}\in\mathcal{S}_{\rho}(T^{\ast}M\times\mathbb{R};\rho^{\ast}\wedge T^{\ast}M)

where βλ\beta_{\lambda} sends (x,ξ,t)TM×(x,\xi,t)\in T^{\ast}M\times\mathbb{R} to (x,λξ,λt)(x,\lambda\xi,\lambda t) and 𝒮ρ\mathcal{S}_{\rho} is the space of Schwartz sections along the fiber of ρ:TM×M\rho:T^{\ast}M\times\mathbb{R}\to M. More precisely, we have

βλ^λm^\displaystyle\beta^{\ast}_{\lambda}\widehat{\mathbb{P}}-\lambda^{m}\widehat{\mathbb{P}} =Iβλu^Iαλ,T(eIt(I)^)λmIu^IT(eIt(I)^)\displaystyle=\sum_{I}\beta_{\lambda}^{\ast}\widehat{u}_{I}\cdot\alpha_{\lambda,\ast}T\left(\widehat{e_{I}t^{\ell(I)}}\right)-\lambda^{m}\sum_{I}\widehat{u}_{I}\cdot T\left(\widehat{e_{I}t^{\ell(I)}}\right)
=Iβλu^Iλ(I)T(eIt(I)^)λmIu^IT(eIt(I)^)\displaystyle=\sum_{I}\beta_{\lambda}^{\ast}\widehat{u}_{I}\cdot\lambda^{\ell(I)}T\left(\widehat{e_{I}t^{\ell(I)}}\right)-\lambda^{m}\sum_{I}\widehat{u}_{I}\cdot T\left(\widehat{e_{I}t^{\ell(I)}}\right)
=Iλ(I)(βλu^Iλm(I)u^I)T(eIt(I)^)𝒮ρ(TM×;ρTM)\displaystyle=\sum_{I}\lambda^{\ell(I)}\left(\beta^{\ast}_{\lambda}\widehat{u}_{I}-\lambda^{m-\ell(I)}\widehat{u}_{I}\right)\cdot T\left(\widehat{e_{I}t^{\ell(I)}}\right)\in\mathcal{S}_{\rho}(T^{\ast}M\times\mathbb{R};\rho^{\ast}\wedge T^{\ast}M)

which implies that βλu^Iλm(I)u^I\beta^{\ast}_{\lambda}\widehat{u}_{I}-\lambda^{m-\ell(I)}\widehat{u}_{I} is of Schwartz class. Thus, uIu_{I} is the Schwartz kernel of some scalar pseudodifferential operator of order m(I)m-\ell(I).

Let Ψm(𝕋M,𝕊)\mathbb{P}\in\Psi^{m}(\mathbb{T}M,\mathbb{S}), as in [EY17], the operator P=|t=1:Cc(M,S)Cc(M,S)P=\mathbb{P}|_{t=1}:C^{\infty}_{c}(M,S)\to C^{\infty}_{c}(M,S) is a properly supported pseudodifferential operator on the spinor bundle SS and all pseudodifferential operators on the spinor bundle can be defined in this way. The full symbol of \mathbb{P} which is defined to be ^\widehat{\mathbb{P}} can be written as

(5.4) ^(x,ξ,t)=pi(x,ξ,t)ωi\widehat{\mathbb{P}}(x,\xi,t)=\sum p_{i}(x,\xi,t)\otimes\omega_{i}

where pi(x,ξ,t)p_{i}(x,\xi,t) is the symbol of scalar properly supported pseudodifferential operator of order mim-i and ωi\omega_{i} is a differential form of order ii. By setting t=1t=1 in (5.4), it justifies the requirement in [BF90, Definition 2.2].

Remark 5.5.

Sometimes, for simplicity of notation, we shall write ~\widetilde{\mathbb{P}} for T(Φ)T(\Phi^{\ast}\mathbb{P}). The essentially homogeneous condition (1.5) becomes

(5.5) α~λ,~λm~C(TM×;ρTM)\widetilde{\alpha}_{\lambda,\ast}\widetilde{\mathbb{P}}-\lambda^{m}\cdot\widetilde{\mathbb{P}}\in C^{\infty}(TM\times\mathbb{R};\rho^{\ast}\wedge T^{\ast}M)

here again, ρ:TM×M\rho:TM\times\mathbb{R}\to M denotes the bundle projection.

As ~\widetilde{\mathbb{P}} belongs to the space of fibered distributions π(TM×,ρTM)\mathcal{E}^{\prime}_{\pi}(TM\times\mathbb{R},\rho^{\ast}\wedge T^{\ast}M) along the fiber of π:TM×M×\pi:TM\times\mathbb{R}\to M\times\mathbb{R}, one may define its derivative with respect to tt as another fibered distribution in π(TM×,ρTM)\mathcal{E}^{\prime}_{\pi}(TM\times\mathbb{R},\rho^{\ast}\wedge T^{\ast}M) given by

t~,f=t~,f~,tf.\langle\partial_{t}\widetilde{\mathbb{P}},f\rangle=\partial_{t}\langle\widetilde{\mathbb{P}},f\rangle-\langle\widetilde{\mathbb{P}},\partial_{t}f\rangle.

Under this light, one can define the Taylor’s expansion of \mathbb{P} to be the series

~|t=0+tt~|t=0+t22t2~|t=0+.\widetilde{\mathbb{P}}|_{t=0}+t\cdot\partial_{t}\widetilde{\mathbb{P}}|_{t=0}+\frac{t^{2}}{2}\cdot\partial_{t}^{2}\widetilde{\mathbb{P}}|_{t=0}+\cdots.

In fact, the class of \mathbb{P} in Ψm(𝕋M,𝕊)/A\Psi^{m}(\mathbb{T}M,\mathbb{S})/A is determined by the Taylor’s expansion of ~\widetilde{\mathbb{P}}.

Proposition 5.6.

If \mathbb{P} and \mathbb{Q} are representatives of some class in Ψm(𝕋M,𝕊)/A\Psi^{m}(\mathbb{T}M,\mathbb{S})/A, then they belong to the same class if and only if they have the same Taylor’s expansion. ∎

Therefore, in the following discussion, we may take the space Ψ(𝕋M,𝕊)/A\Psi^{\ast}(\mathbb{T}M,\mathbb{S})/A as the space of Taylor’s expansion. Moreover, this space has a multiplicative structure: The Taylor’s expansion of \mathbb{P}\ast\mathbb{Q} is given by

~|t=0~|t=0+t(t~|t=0~|t=0+~|t=0t~|t=0)+t22(t2~|t=0~|t=0+t~|t=0t~|t=0+~|t=0t2~|t=0)+.\widetilde{\mathbb{P}}|_{t=0}\ast\widetilde{\mathbb{Q}}|_{t=0}+t\cdot\left(\partial_{t}\widetilde{\mathbb{P}}|_{t=0}\ast\widetilde{\mathbb{Q}}|_{t=0}+\widetilde{\mathbb{P}}|_{t=0}\ast\partial_{t}\widetilde{\mathbb{Q}}|_{t=0}\right)\\ +\frac{t^{2}}{2}\cdot\left(\partial^{2}_{t}\widetilde{\mathbb{P}}|_{t=0}\ast\widetilde{\mathbb{Q}}|_{t=0}+\partial_{t}\widetilde{\mathbb{P}}|_{t=0}\ast\partial_{t}\widetilde{\mathbb{Q}}|_{t=0}+\widetilde{\mathbb{P}}|_{t=0}\ast\partial^{2}_{t}\widetilde{\mathbb{Q}}|_{t=0}\right)+\cdots.

6 Symbols

6.1 Symbol and quantization

Let Ψm(𝕋M,𝕊)\mathbb{P}\in\Psi^{m}(\mathbb{T}M,\mathbb{S}) be a representative of some class in Ψm(𝕋M,𝕊)/A\Psi^{m}(\mathbb{T}M,\mathbb{S})/A, we shall say that the Fourier transformation of ~\widetilde{\mathbb{P}} is a full symbol of the equivalent class []Ψm(𝕋M,𝕊)/A[\mathbb{P}]\in\Psi^{m}(\mathbb{T}M,\mathbb{S})/A. As the Fourier transformation of compactly supported distributions are smooth functions, it is easy to see that the space of full symbols are contained inside the space C(TM×,ρTM)C^{\infty}(T^{\ast}M\times\mathbb{R},\rho^{\ast}\wedge T^{\ast}M). According to [EY17, Proposition 43 and Corollary 45], the full symbol aa is equal to a genuinely homogeneous function at infinity modulo 𝒮ρ(TM×;ρTM)\mathcal{S}_{\rho}(T^{\ast}M\times\mathbb{R};\rho^{\ast}\wedge T^{\ast}M) and satisfies the estimate (1.7).

Of course, this definition of full symbols depend on the choice of representatives, single equivalent class may have more than one full symbols. This ambiguity can be removed by looking at the Taylor’s expansion of full symbols. Indeed, let \mathbb{P}^{\prime} be another representative, the difference \mathbb{P}-\mathbb{P}^{\prime} has zero Taylor’s expansion, and therefore its full symbols also has zero Taylor’s expansion. We denote by SymbmC(TM×;ρTM)\operatorname{Symb}^{m}\subset C^{\infty}(T^{\ast}M\times\mathbb{R};\rho^{\ast}\wedge T^{\ast}M) the space of all full symbols of Ψm(𝕋M,𝕊)\mathbb{P}\in\Psi^{m}(\mathbb{T}M,\mathbb{S}) and by TaylorSymbm\operatorname{Taylor-Symb}^{m} the space of all Taylor’s expansion of full symbols.

Definition 6.1.

Let Ψm(𝕋M,𝕊)\mathbb{P}\in\Psi^{m}(\mathbb{T}M,\mathbb{S}) be a representative, the full symbol map

(6.1) Σ:Ψm(𝕋M,𝕊)/ATaylorSymbm\Sigma:\Psi^{m}(\mathbb{T}M,\mathbb{S})/A\to\operatorname{Taylor-Symb}^{m}

takes the class of \mathbb{P} to the Taylor’s expansion of any full symbols of \mathbb{P}.

Remark 6.2.

There is a more direct way to calculate the full symbol from the distribution. Indeed,

qt(a(x,ξ,t))s(x)\displaystyle q_{t}(a(x,\xi,t))s(x) =qt~,eiYξs(x)\displaystyle=q_{t}\langle\widetilde{\mathbb{P}},e^{iY\cdot\xi}\rangle s(x)
=TxM(x,expx(tY),t)eiYξτ(x,expx(tY),t)s(x)𝑑Y\displaystyle=\int_{T_{x}M}\mathbb{P}(x,\exp_{x}(-tY),t)e^{iY\cdot\xi}\tau(x,\exp_{x}(-tY),t)s(x)dY
=M(x,y,t)eiexpx1yξ/tτ(x,y)s(x)𝑑μ(y)\displaystyle=\int_{M}\mathbb{P}(x,y,t)e^{-i\exp_{x}^{-1}y\cdot\xi/t}\tau(x,y)s(x)d\mu(y)
=t,y(eiexpx1yξ/tsx(y))|y=x\displaystyle=\mathbb{P}_{t,y}(e^{-i\exp_{x}^{-1}y\cdot\xi/t}s_{x}(y))|_{y=x}

where sx(y)=τ(x,y)s(x)s_{x}(y)=\tau(x,y)s(x) and μ\mu is the Riemannian measure on MM. By setting t=1t=1, this recover the formula in [BF90, Definition 2.4].

Example 6.3.

Let DD be the Dirac operator on the spinor bundle SMS\to M, then Lichnerowicz formula shows that

D2=iei2+s/4D^{2}=-\sum_{i}\nabla_{e_{i}}^{2}+s/4

where ss is the scalar curvature. Let 𝔻:C(𝕋M,𝕊)C(M×)\mathbb{D}:C^{\infty}(\mathbb{T}M,\mathbb{S})\to C^{\infty}(M\times\mathbb{R}) be the rr-fibered distribution given by the Schwartz kernel of t2Dt^{2}D on each rr-fiber, namely

𝔻,f(x,t)=t2Df(x,x,t)\left\langle\mathbb{D},f\right\rangle(x,t)=t^{2}Df(x,x,t)

where on the right hand side the Dirac operator is acting on the second variable of ff. Similarly let 𝔻2:C(𝕋M,𝕊)C(M×)\mathbb{D}^{2}:C^{\infty}(\mathbb{T}M,\mathbb{S})\to C^{\infty}(M\times\mathbb{R}) be the rr-fibered distribution given by

𝔻2,f(x,t)=t2D2f(x,x,t)\left\langle\mathbb{D}^{2},f\right\rangle(x,t)=t^{2}D^{2}f(x,x,t)

where the square of Dirac operator D2D^{2} is again, acting on the second variable of ff. The full symbol of 𝔻2\mathbb{D}^{2} is given by a(x,ξ,t)=|ξ|2+s/4t2a(x,\xi,t)=-|\xi|^{2}+s/4\cdot t^{2}.

There is also an inverse to the symbol map Q:TaylorSymbmΨm(𝕋M,𝕊)/AQ:\operatorname{Taylor-Symb}^{m}\to\Psi^{m}(\mathbb{T}M,\mathbb{S})/A which we shall describe now. By definition, a symbol a(x,ξ,t)Symbma(x,\xi,t)\in\operatorname{Symb}^{m} is given by the Fourier transformation of ~\widetilde{\mathbb{P}}. Applying the inverse Fourier transformation, we have

(6.2) ~(x,Y,t)=(2π)nTxMa(x,ξ,t)eiYξ𝑑ξ,\widetilde{\mathbb{P}}(x,Y,t)=(2\pi)^{-n}\int_{T^{\ast}_{x}M}a(x,\xi,t)e^{iY\cdot\xi}d\xi,

applying the map SS on both sides, the equation becomes

(6.3) (x,y,t)=(2π)nτ2(x,y)TxMqt(a(x,ξ,t))eiξexpx1(y)/t𝑑ξ,\mathbb{P}(x,y,t)=(2\pi)^{-n}\tau_{2}(x,y)\int_{T^{\ast}_{x}M}q_{t}\left(a(x,\xi,t)\right)e^{-i\xi\cdot\exp_{x}^{-1}(y)/t}d\xi,

here τ2(x,y)\tau_{2}(x,y) means the parallel translation in the second variable of SSS\boxtimes S^{\ast} from the point xx to yy. Let u¯(y)\overline{u}(y) be the parallel translation of u(y)u(y) to SxS_{x}. As (x,y,1)\mathbb{P}(x,y,1) is the Schwartz kernel of the properly supported pseudodifferential operator P:Cc(M,S)Cc(M,S)P:C^{\infty}_{c}(M,S)\to C^{\infty}_{c}(M,S), we have

(6.4) Pu(x)=(2π)nq(a(x,ξ))eiξexpx1(y)u¯(y)𝑑ξ𝑑yPu(x)=(2\pi)^{-n}\int q(a(x,\xi))e^{-i\xi\cdot\exp_{x}^{-1}(y)}\overline{u}(y)d\xi dy

which recover the formula of [BF90, Equation 2.15].

Definition 6.4.

The quantization map

Q:TaylorSymbΨ(𝕋M,𝕊)/AQ:\operatorname{Taylor-Symb}^{\ast}\to\Psi^{\ast}(\mathbb{T}M,\mathbb{S})/A

sends the class determine by a(x,ξ,t)a(x,\xi,t) to the class determine by the equation (6.3). By construction the quantization map is inverse to the symbol map.

The space TaylorSymb\operatorname{Taylor-Symb}^{\ast} inherit an algebra structure from that of Ψ(𝕋M,𝕊)/A\Psi^{\ast}(\mathbb{T}M,\mathbb{S})/A. Its multiplication formula is particular useful and we shall now describe. Let a,ba,b be two symbols with Taylor’s expansion

atkak(x,ξ)a\sim\sum t^{k}\cdot a_{k}(x,\xi)

and

btkbk(x,ξ),b\sim\sum t^{k}\cdot b_{k}(x,\xi),

their multiplication in TaylorSymbm\operatorname{Taylor-Symb}^{m} is given by the Taylor’s expansion

(6.5) a0#0b0+t(a0#0b1+a1#0b0)+t2(a0#0b2+a1#0b1+a2#0b2)+a_{0}\#_{0}b_{0}+t\left(a_{0}\#_{0}b_{1}+a_{1}\#_{0}b_{0}\right)+t^{2}\left(a_{0}\#_{0}b_{2}+a_{1}\#_{0}b_{1}+a_{2}\#_{0}b_{2}\right)+\cdots

where #0\#_{0} is the multiplication given in Proposition 4.6. With this multiplication formula, the quantization map and the full symbol map are algebra homomorphisms.

If Ψm(𝕋M,𝕊)\mathbb{P}\in\Psi^{m}(\mathbb{T}M,\mathbb{S}) with mm less than n-n, then \mathbb{P} is of trace class. Its supertrace can be calculated as

Str()=(2π)n(2i)n/2TMa(x,ξ,t)𝑑x.\operatorname{Str}(\mathbb{P})=(2\pi)^{-n}\left(\frac{2}{i}\right)^{n/2}\int_{T^{\ast}M}a(x,\xi,t)dx.

Here the full symbol a(x,ξ,t)a(x,\xi,t) take value in TxM\wedge^{\ast}T^{\ast}_{x}M, together with dxdx, the integrand can be integrated against cotangent bundle. Under the light of Taylor’s expansion, the supertrace can be expanded as

Str()tk(2π)nk!(2i)n/2TMtka(x,ξ,0)dx.\operatorname{Str}(\mathbb{P})\sim t^{k}\frac{(2\pi)^{-n}}{k!}\left(\frac{2}{i}\right)^{n/2}\int_{T^{\ast}M}\partial_{t}^{k}a(x,\xi,0)dx.

6.2 Extended symbol space

As mentioned in the Introduction, the symbol space Symbm\operatorname{Symb}^{m} is restrictive in two ways, in this subsection we shall extend the symbol space as well as the quantization map and the full symbol map.

Definition 6.5.

Let SmC(TM×;ρTM)S^{m}\subset C^{\infty}(T^{\ast}M\times\mathbb{R};\rho^{\ast}\wedge T^{\ast}M) be the subspace whose element satisfies the symbol estimate (1.7).

Let aSma\in S^{m}, its Taylor’s expansion is given by

(6.6) a(x,ξ,t)a(x,ξ,0)+tta(x,ξ,0)+t22t2a(x,ξ,0)+a(x,\xi,t)\sim a(x,\xi,0)+t\partial_{t}a(x,\xi,0)+\frac{t^{2}}{2}\partial_{t}^{2}a(x,\xi,0)+\cdots

where the coefficient of tn/n!t^{n}/n! satisfy the relation |xαξβb(x,ξ)|=𝒪((1+|ξ|)mn|β|)|\partial_{x}^{\alpha}\partial_{\xi}^{\beta}b(x,\xi)|=\mathcal{O}\left((1+|\xi|)^{m-n-|\beta|}\right) for all α,β\alpha,\beta. Denote by TaylorSm\operatorname{Taylor-S}^{m} the space of Taylor’s expansion of symbols in SmS^{m}, the quantization map given in Definition 6.4 can be extended to TaylorSm\operatorname{Taylor-S}^{m} by using the formula (6.3) for each aSma\in S^{m}.

Definition 6.6.

Let aSma\in S^{m}, φ\varphi be a cut-off function on the tangent groupoid as in Definition 5.2 and φ(a)𝒟r(𝕋M,𝕊)\mathbb{Q}_{\varphi}(a)\in\mathcal{D}^{\prime}_{r}(\mathbb{T}M,\mathbb{S}) be the following distribution

φ(a)(x,y,t)=(2π)nφ(x,y,t)τ2(x,y)TxMqt(a(x,ξ,t))eiξexpx1(y)/t𝑑ξ.\mathbb{Q}_{\varphi}(a)(x,y,t)=(2\pi)^{-n}\varphi(x,y,t)\tau_{2}(x,y)\int_{T^{\ast}_{x}M}q_{t}(a(x,\xi,t))e^{-i\xi\cdot\exp_{x}^{-1}(y)/t}d\xi.

Let Q(Sm)\operatorname{Q}(S^{m}) be the set of all φ(a)\mathbb{Q}_{\varphi}(a) as φ\varphi varies over all cut-off functions and aa ranges over all SmS^{m}.

It is easy to check that the Taylor’s expansion of φ(a)\mathbb{Q}_{\varphi}(a) is given by

(6.7) b0+tb1+t22b2+b_{0}+tb_{1}+\frac{t^{2}}{2}b_{2}+\cdots

where bi𝒟π(TM,πTM)b_{i}\in\mathcal{D}_{\pi}^{\prime}(TM,\pi^{\ast}\wedge T^{\ast}M) is a fibered distribution along the bundle projection π:TMM\pi:TM\to M given by the oscillatory integral

bi(x,Y)=tka(x,ξ,0)eiYξdξ.b_{i}(x,Y)=\int\partial_{t}^{k}a(x,\xi,0)e^{iY\cdot\xi}d\xi.

Let TaylorQ(Sm)\operatorname{Taylor-Q(S^{m})} be the space of Taylor’s expansion of elements in Q(Sm)Q(S^{m}).

Definition 6.7.

The extended symbol map

Σ:TaylorQ(Sm)TaylorSm\Sigma:\operatorname{Taylor-Q(S^{m})}\to\operatorname{Taylor-S^{m}}

is given by sending (6.7) to (6.6).

It is not clear, in our context, if the space Q(Sm)Q(S^{m}) has an algebra structure, however, as bib_{i}’s are Schwartz kernel of classical pseudodifferential operators on tangent space, the algebra structure of TaylorQ(Sm)\operatorname{Taylor-Q(S^{m})} is clear. Let φ,φ\varphi,\varphi^{\prime} be two cut-off functions, the difference φ(a)φ(a)=φφ(a)\mathbb{Q}_{\varphi}(a)-\mathbb{Q}_{\varphi^{\prime}}(a)=\mathbb{Q}_{\varphi-\varphi^{\prime}}(a) has zero Taylor’s expansion at t=0t=0.

Definition 6.8.

The extended quantization map is a map

Q:TaylorSTaylorQ(S)Q:\operatorname{Taylor-S}^{\ast}\to\operatorname{Taylor-Q(S^{\ast})}

that sends the Taylor’s expansion of any aSma\in S^{m} to the Taylor’s expansion of φ(a)\mathbb{Q}_{\varphi}(a) for any cut-off function φ\varphi.

Proposition 6.9.

Let akC(TM,πTM)a_{k}\in C^{\infty}(TM,\pi^{\ast}\wedge T^{\ast}M) be a sequence of smooth section that satisfy

xαξβak(x,ξ)=𝒪(1+|ξ|)mk|β|\partial_{x}^{\alpha}\partial_{\xi}^{\beta}a_{k}(x,\xi)=\mathcal{O}(1+|\xi|)^{m-k-|\beta|}

for all α,β\alpha,\beta, then there exist a(x,ξ,t)Sma(x,\xi,t)\in S^{m} which has Taylor’s expansion a(x,ξ,t)tkak(x,ξ)a(x,\xi,t)\sim\sum t^{k}a_{k}(x,\xi).

Proof.

(see [BF90, Lemma 3.11].) Choose ϕC()\phi\in C^{\infty}(\mathbb{R}) such that ϕ(x)=0\phi(x)=0 if x1x\leq 1 and ϕ(x)=1\phi(x)=1 if x2x\geq 2. Let Ck=max|α|,|β|ksup|(1+|ξ|)|β|mxαξβak(x,ξ)|C_{k}=\max_{|\alpha|,|\beta|\leq k}\sup\left|(1+|\xi|)^{|\beta|-m}\partial_{x}^{\alpha}\partial_{\xi}^{\beta}a_{k}(x,\xi)\right| and choose a decreasing sequence of positive numbers εk\varepsilon_{k} that converges to zero and satisfy

εk(k!2iCk)2.\varepsilon_{k}\leq\left(k!\cdot 2^{i}\cdot C_{k}\right)^{-2}.

Write ϕk(x)\phi_{k}(x) for ϕ(εkx)\phi(\varepsilon_{k}x) and set

(6.8) a(x,ξ,t)=k=0tkϕk((t|ξ|+t)2)ak(x,ξ).a(x,\xi,t)=\sum_{k=0}^{\infty}t^{k}\phi_{k}((t|\xi|+t)^{-2})a_{k}(x,\xi).

Notice that for any nonzero tt there is a NN\in\mathbb{N} such that εk(t|ξ|+t)2<1\varepsilon_{k}(t|\xi|+t)^{-2}<1 for all k>Nk>N. Therefore, the summation (6.8) is locally finite and well-defined. As for the smoothness at t=0t=0, we observe that the differential of (6.8) at any (x,ξ,t)TM×(x,\xi,t)\in T^{\ast}M\times\mathbb{R}^{\ast} is given by

(6.9) xαξβtγa(x,ξ,t)=k=0Cγ1γ2Cβ1β2k!tk|γ1|ξβ2tγ2(ϕk((t|ξ|+t)2))xαξβ1ak(x,ξ).\partial_{x}^{\alpha}\partial_{\xi}^{\beta}\partial_{t}^{\gamma}a(x,\xi,t)=\sum_{k=0}^{\infty}C_{\gamma_{1}\gamma_{2}}C_{\beta_{1}\beta_{2}}\cdot k!\cdot t^{k-|\gamma_{1}|}\partial_{\xi}^{\beta_{2}}\partial_{t}^{\gamma_{2}}\left(\phi_{k}((t|\xi|+t)^{-2})\right)\partial_{x}^{\alpha}\partial_{\xi}^{\beta_{1}}a_{k}(x,\xi).

The sum k=0\sum_{k=0}^{\infty} can be split into two parts k|α|+|β|+|γ|+1\sum_{k\leq|\alpha|+|\beta|+|\gamma|+1} and k|α|+|β|+|γ|+2\sum_{k\geq|\alpha|+|\beta|+|\gamma|+2} where the first part is finite and clearly continuous at t=0t=0.

For the second part we may assume that εk(t|ξ|+t)21\varepsilon_{k}(t|\xi|+t)^{-2}\geq 1, or equivalently, tεk(|ξ|+1)1εkt\leq\sqrt{\varepsilon_{k}}(|\xi|+1)^{-1}\leq\sqrt{\varepsilon_{k}}, and the factor t|γ2|ξβ2tγ2(ϕk((t|ξ|+t)2))t^{|\gamma_{2}|}\partial_{\xi}^{\beta_{2}}\partial_{t}^{\gamma_{2}}\left(\phi_{k}((t|\xi|+t)^{-2})\right) is controlled by (1+|ξ|)|β2|(1+|\xi|)^{-|\beta_{2}|} times a constant independent of t,kt,k and |ξ||\xi|. So, the absolute value of each term in the sum k|α|+|β|+|γ|+2\sum_{k\geq|\alpha|+|\beta|+|\gamma|+2} is controlled by a constant that only depends on α,β,γ\alpha,\beta,\gamma times k!tk|γ|Ck(1+|ξ|)m|β|k.k!t^{k-|\gamma|}C_{k}\cdot(1+|\xi|)^{m-|\beta|-k}. And the choice of εk\varepsilon_{k} ensures that each summand in the second part is controlled by t2k(1+|ξ|)mβ|kt\cdot 2^{-k}\cdot(1+|\xi|)^{m-\beta|-k} This proves the smoothness as well as the symbol estimate (1.7).

Now, we shall verify the Taylor’s expansion condition. Indeed, since ϕk((t|ξ|+t)2)1\phi_{k}((t|\xi|+t)^{-2})-1 vanishes to infinite order at t=0t=0, by setting α,β=0\alpha,\beta=0 in (6.9), we have tγa(x,ξ,0)=aγ(x,ξ)\partial_{t}^{\gamma}a(x,\xi,0)=a_{\gamma}(x,\xi). This completes the proof.

The space TaylorS\operatorname{Taylor-S}^{\ast} can be given a multiplication formula by the formula (6.5). In fact, this multiplication formula turn TaylorS\operatorname{Taylor-S}^{\ast} into a graded algebra. Indeed, let a,bC(TM,πTM)a,b\in C^{\infty}(TM,\pi^{\ast}\wedge T^{\ast}M) satisfy xαξβa(x,ξ)=𝒪(1+|ξ|)m|β|\partial_{x}^{\alpha}\partial_{\xi}^{\beta}a(x,\xi)=\mathcal{O}(1+|\xi|)^{m-|\beta|} and xαξβb(x,ξ)=𝒪(1+|ξ|)n|β|\partial_{x}^{\alpha}\partial_{\xi}^{\beta}b(x,\xi)=\mathcal{O}(1+|\xi|)^{n-|\beta|} for all α,β\alpha,\beta. Then

a#0b(x,ξ)=exp(12κ(ξ,η))a(x,ξ)b(x,η)|ξ=ηa\#_{0}b(x,\xi)=\exp\left(-\frac{1}{2}\kappa(\partial_{\xi},\partial_{\eta})\right)a(x,\xi)b(x,\eta)|_{\xi=\eta}

can be expanded to a finite sum where each term is some derivatives with respect to ξ\xi or η\eta. Its highest order term is a(x,ξ)b(x,ξ)a(x,\xi)b(x,\xi) which satisfy

xαξβ(a(x,ξ)b(x,ξ))=𝒪(1+|ξ|)m+n|β|.\partial_{x}^{\alpha}\partial_{\xi}^{\beta}\left(a(x,\xi)b(x,\xi)\right)=\mathcal{O}(1+|\xi|)^{m+n-|\beta|}.

Therefore according to the Proposition 6.9, the multiplication of aTaylorSma\in\operatorname{Taylor-S}^{m} and bTaylorSnb\in\operatorname{Taylor-S}^{n} belongs to TaylorSm+n\operatorname{Taylor-S}^{m+n}.

Proposition 6.10.

The extended full symbol map is an algebra isomorphism whose inverse is given by the extended quantization map.

Proof.

Let tkk\sum t^{k}\mathbb{P}_{k} and tkk\sum t^{k}\mathbb{Q}_{k} be two Taylor’s expansions in TaylorQ(S)\operatorname{Taylor-Q(S^{\ast})} whose extended full symbol are tkak\sum t^{k}a_{k} and tkbk\sum t^{k}b_{k} respectively, their multiplication is given by

00+t(10+01)+t22(20+211+02)+\mathbb{P}_{0}\ast\mathbb{Q}_{0}+t\left(\mathbb{P}_{1}\ast\mathbb{Q}_{0}+\mathbb{P}_{0}\ast\mathbb{Q}_{1}\right)+\frac{t^{2}}{2}\left(\mathbb{P}_{2}\ast\mathbb{Q}_{0}+2\mathbb{P}_{1}\ast\mathbb{Q}_{1}+\mathbb{P}_{0}\ast\mathbb{Q}_{2}\right)+\cdots

whose Fourier transformation according to Corollary 4.7 is given by

(6.10) a0#0b0+t(a1#0b0+a0#0b1)+t22(a2#0b0+2a1#0b1+a0#0b2)+a_{0}\#_{0}b_{0}+t(a_{1}\#_{0}b_{0}+a_{0}\#_{0}b_{1})+\frac{t^{2}}{2}(a_{2}\#_{0}b_{0}+2a_{1}\#_{0}b_{1}+a_{0}\#_{0}b_{2})+\cdots

which coincide with the algebra structure of TaylorS\operatorname{Taylor-S^{\ast}}. ∎

7 Index theory

Consider the differential equation

(7.1) fτ+𝔻2f=0.\frac{\partial f}{\partial\tau}+\mathbb{D}^{2}f=0.

When restricting on each t0t\neq 0 fiber on the tangent groupoid, the fundamental solution, namely the solution with initial condition f|τ=0=δf|_{\tau=0}=\delta, is the heat kernel eτt2D2(x,y)C(M×M,SS)e^{-\tau t^{2}D^{2}}(x,y)\in C^{\infty}(M\times M,S\boxtimes S^{\ast}). An important problem is that for fixed τ0\tau\neq 0 whether this solution extend smoothly to t=0t=0 as a smooth section of the rescaled bundle.

Assume that the solution ff belongs to Q(S)Q(S^{\ast}), and whose full symbol has Taylor’s expansion,

a0τ(x,ξ)+t22a2τ(x,ξ)+t424a4τ(x,ξ)+,a^{\tau}_{0}(x,\xi)+\frac{t^{2}}{2}a^{\tau}_{2}(x,\xi)+\frac{t^{4}}{24}a^{\tau}_{4}(x,\xi)+\cdots,

then, according to the Proposition 6.10, the full symbol of 𝔻2f\mathbb{D}^{2}f has asymptotic expansion

|ξ|2#0a0τ(x,ξ)+t22(|ξ|2#0a2τ(x,ξ)+s2a0τ(x,ξ))+-|\xi|^{2}\#_{0}a^{\tau}_{0}(x,\xi)+\frac{t^{2}}{2}\left(-|\xi|^{2}\#_{0}a^{\tau}_{2}(x,\xi)+\frac{s}{2}a^{\tau}_{0}(x,\xi)\right)+\cdots

where the convolution #0\#_{0} is given by the formula (4.4). As |ξ|2|\xi|^{2} is a polynomial of order 22, the convolution |ξ|2#0a2kτ-|\xi|^{2}\#_{0}a^{\tau}_{2k} has only three terms and can be explicitly calculated as

|ξ|2#0a2kτ(x,ξ)=|ξ|2a2kτ(x,ξ)+κ(ei,ej)ξiξja2kτ(x,ξ)+14κ(ei,ep)κ(ep,ej)ξiξja2kτ(x,ξ).-|\xi|^{2}\#_{0}a_{2k}^{\tau}(x,\xi)=-|\xi|^{2}a_{2k}^{\tau}(x,\xi)+\kappa(e_{i},e_{j})\xi_{i}\partial_{\xi_{j}}a_{2k}^{\tau}(x,\xi)\\ +\frac{1}{4}\kappa(e_{i},e_{p})\kappa(e_{p},e_{j})\partial_{\xi_{i}}\partial_{\xi_{j}}a_{2k}^{\tau}(x,\xi).

At the level of full symbols, the differential equation (7.1) becomes

a2kττ(x,ξ)+(|ξ|2κ(ξ,ξ)14κκ(ξ,ξ))a2kτ(x,ξ)+s(k1)(k12)a2k2τ(x,ξ)=0.\frac{\partial a_{2k}^{\tau}}{\partial\tau}(x,\xi)+\left(|\xi|^{2}-\kappa(\xi,\partial_{\xi})-\frac{1}{4}\kappa\wedge\kappa(\partial_{\xi},\partial_{\xi})\right)a_{2k}^{\tau}(x,\xi)\\ +s\left(k-1\right)\left(k-\frac{1}{2}\right)a^{\tau}_{2k-2}(x,\xi)=0.

with initial condition a2kτ|τ=0=0a^{\tau}_{2k}|_{\tau=0}=0 for all k1k\geq 1 and

(7.2) a0ττ(x,ξ)+(|ξ|2κ(ξ,ξ)14κκ(ξ,ξ))a0τ(x,ξ)=0,\frac{\partial a_{0}^{\tau}}{\partial\tau}(x,\xi)+\left(|\xi|^{2}-\kappa(\xi,\partial_{\xi})-\frac{1}{4}\kappa\wedge\kappa(\partial_{\xi},\partial_{\xi})\right)a_{0}^{\tau}(x,\xi)=0,

with initial condition a0τ|τ=0=1a_{0}^{\tau}|_{\tau=0}=1. The fundamental solution Aτ(ξ,η)A^{\tau}(\xi,\eta) of (7.2) is given by the Mehler’s formula which is Schwartz in ξ\xi and one can solve a2kτ(x,ξ)a_{2k}^{\tau}(x,\xi) recursively by the formula

a2kτ(x,ξ)=s(k1)(k12)0τAττ(ξ,η)a2k2τ(x,η)𝑑τ𝑑ηa_{2k}^{\tau}(x,\xi)=-s\left(k-1\right)\left(k-\frac{1}{2}\right)\int_{0}^{\tau}A^{\tau-\tau^{\prime}}(\xi,\eta)a^{\tau^{\prime}}_{2k-2}(x,\eta)d\tau^{\prime}d\eta

which is also Schwartz in ξ\xi.

According to Proposition 6.9, there is a full symbol a(x,ξ,t)a(x,\xi,t) which has asymptotic expansion

(7.3) a0τ(x,ξ)+t22a2τ(x,ξ)+t424a4τ(x,ξ)+.a^{\tau}_{0}(x,\xi)+\frac{t^{2}}{2}a^{\tau}_{2}(x,\xi)+\frac{t^{4}}{24}a^{\tau}_{4}(x,\xi)+\cdots.

It is easy to see that the full symbol a(x,ξ,t)a(x,\xi,t) is an approximate solution to the equation (7.1), namely

aτ(x,ξ,t)(|ξ|2+s4t2)0a(x,ξ,t)\frac{\partial a}{\partial\tau}(x,\xi,t)-\left(-|\xi|^{2}+\frac{s}{4}t^{2}\right)\ast_{0}a(x,\xi,t)

vanishes to infinite order at t=0t=0. By passing to quantization, we get an approximate solution ff^{\prime} to (7.1).

Proposition 7.1.

Let

g=fτ+𝔻2fg=\frac{\partial f^{\prime}}{\partial\tau}+\mathbb{D}^{2}f^{\prime}

be the smooth section of the rescaled bundle that vanishes to infinite order at t=0t=0, then the differential equation

fτ+𝔻2f=g\frac{\partial f}{\partial\tau}+\mathbb{D}^{2}f=g

with the initial condition f|τ=0=0f|_{\tau=0}=0 has solution belongs to C(𝕋M,𝕊)C^{\infty}(\mathbb{T}M,\mathbb{S}).

Proof.

For any nonzero tt, the equation can be solved by the formula

f(τ)(x,y,t)=0τet2(ττ)D2(x,y)g(τ)(x,y,t)𝑑τ.f(\tau)(x,y,t)=\int_{0}^{\tau}e^{-t^{2}(\tau-\tau^{\prime})D^{2}}(x,y)g(\tau^{\prime})(x,y,t)d\tau^{\prime}.

Then as gg vanishes to infinite order at t=0t=0, the section f(τ)C(M×M×;SS)f(\tau)\in C^{\infty}(M\times M\times\mathbb{R}^{\ast};S^{\ast}\boxtimes S) vanishes to infinite order as t0t\to 0. The section f(τ)f(\tau) can be locally written as a finite sum

f(τ)(x,y,t)=tnfi(x,y,t)si(x,y)f(\tau)(x,y,t)=\sum t^{n}f_{i}(x,y,t)s_{i}(x,y)

where fif_{i} are smooth functions on M×M×M\times M\times\mathbb{R}^{\ast} that vanishes as t0t\to 0, sis_{i} are sections of SSM×MS^{\ast}\boxtimes S\to M\times M and nn is the dimension of MM. Then each fif_{i} can be extended to a smooth function on the tangent groupoid and si(x,y)tns_{i}(x,y)t^{n} is an element of the module S(𝕋M)S(\mathbb{T}M) defined in Definition 3.2 and is a smooth section of the rescaled bundle. Overall, f(τ)f(\tau) can be extended to the smooth section of the rescaled bundle 𝕊\mathbb{S}. ∎

We see that fff-f^{\prime} is a genuine solution to the heat equation (7.1) and it has the same Taylor’s expansion as the approximate solution ff^{\prime}. Therefore, the image of (7.3) under the extended quantization map is precisely the asymptotic expansion of the heat kernel.

References

  • [BF90] Jonathan Block and Jeffrey Fox. Asymptotic pseudodifferential operators and index theory. In Geometric and topological invariants of elliptic operators (Brunswick, ME, 1988), volume 105 of Contemp. Math., pages 1–32. Amer. Math. Soc., Providence, RI, 1990.
  • [EY17] Erik van Erp and Robert Yuncken. A groupoid approach to pseudodifferential calculii. J. Reine Angew. Math., Published Online, 2017. doi.org/10.1515/crelle-2017-0035.
  • [Get83] Ezra Getzler. Pseudodifferential operators on supermanifolds and the Atiyah-Singer index theorem. Comm. Math. Phys., 92(2):163–178, 1983.
  • [H0̈3] Lars Hörmander. The analysis of linear partial differential operators. I. Classics in Mathematics. Springer-Verlag, Berlin, 2003. Distribution theory and Fourier analysis, Reprint of the second (1990) edition [Springer, Berlin; MR1065993 (91m:35001a)].
  • [HSSH18] Ahmad Reza Haj Saeedi Sadegh and Nigel Higson. Euler-like vector fields, deformation spaces and manifolds with filtered structure. Doc. Math., 23:293–325, 2018.
  • [HY19] Nigel Higson and Zelin Yi. Spinors and the tangent groupoid. Doc. Math., 24:1677–1720, 2019.
  • [LMV17] Jean-Marie Lescure, Dominique Manchon, and Stéphane Vassout. About the convolution of distributions on groupoids. J. Noncommut. Geom., 11(2):757–789, 2017.
  • [MS67] H. P. McKean, Jr. and I. M. Singer. Curvature and the eigenvalues of the Laplacian. J. Differential Geometry, 1(1):43–69, 1967.

School of Mathematical Sciences, Fudan University, Shanghai, 200433, China

School of Mathematical Sciences, Tongji University, Shanghai, 200092, China