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

A proof of a trace formula by Richard Melrose

Yves Colin de Verdière111Université Grenoble-Alpes, Institut Fourier, Unité mixte de recherche CNRS-UGA 5582, BP 74, 38402-Saint Martin d’Hères Cedex (France); [email protected]

The goal of this note is to give a new proof of the wave trace formula proved by Richard Melrose in the impressive paper [Me-84]. This trace formula is an extension of the Chazarain-Duistermaat-Guillemin trace formula (denoted “CDG trace formula” in this paper) to the case of a sub-Riemannian (“sR”) Laplacian on a 3D contact closed manifold. The proof uses a normal form constructed in the papers [CHT-18, CHT-21], following the pioneering work [Me-84], in order to reduce to the case of the invariant Laplacian on the 3D-Heisenberg group. We need also the propagation of singularities results of Victor Ivrii, Bernard Lascar and Richard Melrose [Iv-76, La-82, Me-86].

Acknowledgments: many thanks to Cyril for very useful comments!

1 The CDG trace formula

The following result was proved by Chazarain [Ch-74] and refined by Duistermaat and Guillemin [DG-75] (see also [CdV-73, CdV-07] and Appendix F for the history of the trace formulae):

Theorem 1.1

Let (M,g)(M,g) be a closed connected smooth Riemannian manifold and λ1=0<λ2\lambda_{1}=0<\lambda_{2}\leq\cdots the spectrum of the Laplace operator, then we have the following equality of Schwartz distributions:

j=1eitλj=T0(t)+γ𝒫Tγ(t)modC\sum_{j=1}^{\infty}e^{it\sqrt{\lambda_{j}}}=T_{0}(t)+\sum_{\gamma\in{\cal P}}T_{\gamma}(t)~{}{\rm mod}~{}C^{\infty}

where 𝒫{\cal P} is the set of periodic geodesics, SingSupp(T0)={0}{\rm SingSupp}(T_{0})=\{0\} and, for γ\gamma a periodic geodesic, SingSupp(Tγ){Lγ}{Lγ}{\rm SingSupp}(T_{\gamma})\subset\{L_{\gamma}\}\cup\{-L_{\gamma}\}, where LγL_{\gamma} is the length of γ\gamma. Moreover, if γ\gamma is non degenerate,

Tγ(t)=L0eiπm(γ)/2πdet(IdPγ)12(t+i0Lγ)1(1+j=1aj(tLγ)j)T_{\gamma}(t)=\frac{L_{0}e^{i\pi m(\gamma)/2}}{\pi{\rm det}({\rm Id}-P_{\gamma})^{\frac{1}{2}}}(t+i0-L_{\gamma})^{-1}\left(1+\sum_{j=1}^{\infty}a_{j}(t-L_{\gamma})^{j}\right)

where

  • L0L_{0} is the length of the primitive geodesic associated to γ\gamma

  • m(γ)m(\gamma) is the Morse index of γ\gamma

  • PγP_{\gamma} is the linearized Poincaré map.

This result gives a nicer proof of the main result of my thesis [CdV-73] saying that, in the generic case, the Length spectrum, i.e. the set of lengths of closed geodesics, is a spectral invariant. This formula holds for any elliptic self-adjoint pseudo-differential operator PP of degree 11 by replacing the periodic geodesics by the periodic orbits of the Hamiltonian flow of the principal symbol of PP and the Morse index by a Maslov index. Our goal is to prove that the same statement holds for sR Laplacians on a closed contact 3-manifold. What we mean here by ‘Melrose’s trace formula’ is the formulae of theorem 1, interpreted in this contact sR setting. Richard Melrose gave a proof which I found difficult (see Theorem 6.4 in [Me-84]).

2 Review of basic facts and notations

For more details on this section, one can look at the paper [CHT-18].

2.1 Contact 3D sR manifolds

In what follows, MM is a closed (compact without boundary) connected manifold of dimension 33 equipped with a smooth volume form |dq||dq|. We consider also an oriented contact distribution globally defined as the kernel of a non vanishing real valued 1-form α\alpha so that αdα\alpha\wedge d\alpha is a volume form. Let also gg be a metric on the distribution D=kerαD=\ker\alpha. The “co-metric” gg^{\star} is defined by g(q,p):=p|Dq2g^{\star}(q,p):=\|p_{|D_{q}}\|^{2} where the norm is the dual norm of g(q)g(q). To such a set of data is associated

  • A geodesic flow denoted by Gt,tG_{t},~{}t\in{\mathbb{R}}: the Hamiltonian flow of g\sqrt{g^{\star}}. The geodesics are projections of the orbits of that flow onto MM and are everywhere tangent to DD. We will often prefer to consider the geodesic flow as the restriction to g=1g^{\star}=1 of the Hamiltonian flow of 12g\frac{1}{2}g^{\star}.

  • A Laplacian which is locally given by Δ=X1X1+X2X2\Delta=X_{1}^{\star}X_{1}+X_{2}^{\star}X_{2} where (X1,X2)(X_{1},X_{2}) is an orthonormal frame of DD and the adjoint is taken with respect to the measure |dq||dq|.

  • A canonical choice of a 1-form αg\alpha_{g} defining DD by assuming that dαgd\alpha_{g} restricted to DD is the oriented gg-volume form on DD.

From a famous theorem of Hörmander, we know the Laplacian is sub-elliptic and hence has a compact resolvent and a discrete spectrum λ1=0<λ2\lambda_{1}=0<\lambda_{2}\leq\cdots with smooth eigenfunctions. It follows from the sR Weyl law

#{j|λjλ}Cλ2\#\{j|\lambda_{j}\leq\lambda\}\sim C\lambda^{2}

that

Trace(eitΔ)=j=1eitλj{\rm Trace}(e^{it\sqrt{\Delta}})=\sum_{j=1}^{\infty}e^{it\sqrt{\lambda_{j}}}

is a well defined Schwartz distribution often called the wave trace (see Appendix A for the link with the wave equation). Our goal is to extend the CDG formula to this case.

The form αg\alpha_{g} defines a Reeb vector field R\vec{R} by the equations αg(R)=1\alpha_{g}(\vec{R})=1 and ι(R)dαg=0\iota(\vec{R})d\alpha_{g}=0. This vector field admits the following Hamiltonian interpretation: the cone Σ=D\Sigma=D^{\perp} is a symplectic sub-cone of TM0T^{\star}M\setminus 0 222if Σ={(q,tα(q)|t0,qM}\Sigma=\{(q,t\alpha(q)|t\in{\mathbb{R}}\setminus 0,~{}q\in M\}, the symplectic form on TMT^{\star}M restricts to tdα+dtαtd\alpha+dt\wedge\alpha whose square is the volume form tdtαdα-tdt\wedge\alpha\wedge d\alpha. We define the Hamiltonian ρ:Σ\rho:\Sigma\rightarrow{\mathbb{R}} by ρ(α)=α/αg\rho(\alpha)=\alpha/\alpha_{g} where αΣ\alpha\in\Sigma is a covector vanishing on DD. The Hamiltonian vector field of ρ\rho is homogeneous of degree 0 and the projection of this field onto MM is the Reeb vector field R\vec{R} (see section 2.4 of [CHT-18]).

2.2 The 3D Heisenberg group H3H_{3}

For this section, the reader could look at Section 3.1 of [CHT-18]. We identify the Heisenberg group H3H_{3} with x,y,z3{\mathbb{R}}^{3}_{x,y,z} with the group law

(x,y,z)(x,y,z)=(x+x,y+y,z+z+12(xyyx))(x,y,z)\star(x^{\prime},y^{\prime},z^{\prime})=(x+x^{\prime},y+y^{\prime},z+z^{\prime}+\frac{1}{2}(xy^{\prime}-yx^{\prime}))

and the Lie algebra is generated by X,Y,ZX,Y,Z with

X=x+12yz,Y=y12xz,Z=zX=\partial_{x}+\frac{1}{2}y\partial_{z},~{}Y=\partial_{y}-\frac{1}{2}x\partial_{z},~{}Z=\partial_{z}

We choose (D,g)(D,g) by asking that (X,Y)(X,Y) is an oriented orthonormal basis of DD for gg and |dq|=|dxdydz||dq|=|dxdydz|. We have [X,Y]=Z[X,Y]=-Z and the Reeb vector field is ZZ. The Laplacian is Δ3=(X2+Y2)\Delta_{3}=-(X^{2}+Y^{2}) and can be rewritten as

Δ3=|Z|((XZ)2+(YZ)2)\Delta_{3}=-|Z|\left(\left(\frac{X}{\sqrt{Z}}\right)^{2}+\left(\frac{Y}{\sqrt{Z}}\right)^{2}\right) (1)

on the complement of the kernel of ZZ. We write this as

Δ3=|Z|Ω\Delta_{3}=|Z|\Omega

where Ω\Omega is an harmonic oscillator with spectrum {2l+1|l=0,1,}\{2l+1|l=0,1,\cdots\} (see [CHT-18], prop. 3.1).

We will need the following “confining” result (see [Le-20]):

Lemma 2.1

Given T>0T>0 , σ0Σ0\sigma_{0}\in\Sigma\setminus 0 and UU a conic neighbourhood of σ0\sigma_{0}, there exists a conic neighbourhood VUV\subset\subset U of σ0\sigma_{0} so that t[T,T]\forall t\in[-T,T], Gt(V)UG_{t}(V)\subset U.

3 Speed of propagation

First, we have the following

Theorem 3.1

If uu is a solution of an sR wave equation,

t,Support(u(t))B(Support(u(t=0))Support(du/dt(t=0)),|t|)\forall t\in{\mathbb{R}},~{}{\rm Support}(u(t))\subset B({\rm Support}(u(t=0))\cup{\rm Support}(du/dt(t=0)),|t|)

where B(A,r)B(A,r) is the closed sR neighbourhood  of radius rr of AA.

This result follows from the Riemannian case by passing to the limits (see Section 3 of [Me-86]).

We will also use the following Theorem due to Victor Ivrii, Bernard Lascar and Richard Melrose [Iv-76, La-82, Me-86] and revisited by Cyril Letrouit [Le-21]):

Theorem 3.2

If e(t,q,q)e(t,q,q^{\prime}) is the wave kernel of a sR Laplacian whose characteristic manifold Σ\Sigma is symplectic, i.e. ee is the Schwartz kernel of cos(tΔ)\cos(t\sqrt{\Delta}), then

WF(e){(q,p,q,p,t,τ)|τ=±g,(q,p)=G±t(q,p)}{(q,p,q,p,t,0)}WF^{\prime}(e)\subset\{(q,p,q^{\prime},p^{\prime},t,\tau)|\tau=\pm\sqrt{g^{\star}},(q,p)=G_{\pm t}(q^{\prime},p^{\prime})\}\cup\{(q,p,q,p,t,0)\}

where GtG_{t} is the geodesic flow.

4 The local wave trace for the Heisenberg group

As a preparation, we will prove the local trace formula for the 3D-Heisenberg group H3H_{3}. The Laplacian Δ3\Delta_{3} commutes with ZZ. We can hence use a partial Fourier decomposition of L2(H3)L^{2}(H_{3}) identifying it with the Hilbert integral

L2(3)=ζ𝑑ζL^{2}({\mathbb{R}}^{3})=\int_{\mathbb{R}}^{\oplus}{\cal H}_{\zeta}d\zeta

where ζ:={f|f(x,y,z+a)=eiaζf(x,y,z)}{\cal H}_{\zeta}:=\{f|f(x,y,z+a)=e^{ia\zeta}f(x,y,z)\} which is identified to L2(2)L^{2}({\mathbb{R}}^{2}) by looking at the value of ff at z=0z=0. In what follows, we will omit the space 0{\cal H}_{0} which corresponds to a flat 2D-Euclidian Laplacian. Using this decomposition, the Laplacian rewrites as follows:

Δ3=l=0(2l+1)|ζ|Kζl𝑑ζ\Delta_{3}=\sum_{l=0}^{\infty}(2l+1)\int_{\mathbb{R}}^{\oplus}|\zeta|K_{\zeta}^{l}d\zeta

where the operator KζlK_{\zeta}^{l} is the projector on the ll-th Landau level of Δζ\Delta_{\zeta}, the restriction of Δ3\Delta_{3} to ζ{\cal H}_{\zeta}, which is a magnetic Schrödinger operator

Hζ:=((x+iζy/2)2+(yiζx/2)2)H_{\zeta}:=-\left((\partial_{x}+i\zeta y/2)^{2}+(\partial_{y}-i\zeta x/2)^{2}\right)

on 2{\mathbb{R}}^{2} with magnetic field ζdxdy\zeta dx\wedge dy. The Schwartz kernel of KζlK_{\zeta}^{l} satisfies

Kζl(m,m)=|ζ|2πK_{\zeta}^{l}(m,m)=\frac{|\zeta|}{2\pi}

(see Appendix B).

Hence the half-wave operator writes

eitΔ3=l=0eit(2l+1)|ζ|Kζl𝑑ζe^{it\sqrt{\Delta_{3}}}=\sum_{l=0}^{\infty}\int_{\mathbb{R}}^{\oplus}e^{it\sqrt{(2l+1)|\zeta|}}K_{\zeta}^{l}d\zeta

and the local distributional trace is given by

Trace(eitΔ3f)=12π(3f(q)|dq|)l=0eit(2l+1)|ζ||ζ|𝑑ζ{\rm Trace}\left(e^{it\sqrt{\Delta_{3}}}f\right)=\frac{1}{2\pi}\left(\int_{{\mathbb{R}}^{3}}f(q)|dq|\right)\sum_{l=0}^{\infty}\int_{\mathbb{R}}e^{it\sqrt{(2l+1)|\zeta|}}|\zeta|d\zeta

for fC0(H3)f\in C_{0}^{\infty}(H_{3}). This trace can be explicitly computed by using the distributional Fourier transform

0eiτuu3𝑑u=6(τ+i0)4.\int_{0}^{\infty}e^{i\tau u}u^{3}du=6(\tau+i0)^{-4}.

We get, for t0t\neq 0,

Trace(eitΔ3f)=6πt4(l=01(2l+1)2)H3f|dq|.{\rm Trace}\left(e^{it\sqrt{\Delta_{3}}}f\right)=\frac{6}{\pi t^{4}}\left(\sum_{l=0}^{\infty}\frac{1}{(2l+1)^{2}}\right)\int_{H_{3}}f|dq|.

In particular, this trace is smooth outside t=0t=0 which is consistent with the fact that there is no periodic geodesic in H3H_{3}.

The same result holds for the trace formula microlocalized near Σ\Sigma:

Proposition 4.1

Let PP be a pseudo-differential operator  of degree 0, which is compactly supported and so that WF(P){ζ=0}=WF^{\prime}(P)\cap\{\zeta=0\}=\emptyset. Then

Trace(eitΔ3P){\rm Trace}\left(e^{it\sqrt{\Delta_{3}}}P\right)

is smooth outside t=0t=0.

Proof.– We will first prove that

Trace(eitΔ3χ(Δ3|Z|2,q)),{\rm Trace}\left(e^{it\sqrt{\Delta_{3}}}\chi\left(\frac{\Delta_{3}}{|Z|^{2}},q\right)\right)~{},

with χC0(×H3)\chi\in C_{0}^{\infty}({\mathbb{R}}\times H_{3}), which rewrites as

12πl=0H3|dq|eit(2l+1)|ζ|χ(2l+1|ζ|,q)|ζ|𝑑ζ,\frac{1}{2\pi}\sum_{l=0}^{\infty}\int_{H_{3}}|dq|\int_{\mathbb{R}}e^{it\sqrt{(2l+1)|\zeta|}}\chi\left(\frac{2l+1}{|\zeta|},q\right)|\zeta|d\zeta~{},

is smooth outside t=0t=0. Let us define

Il(t)=eit(2l+1)|ζ|χ(2l+1|ζ|,q)|ζ|𝑑ζ.I_{l}(t)=\int_{\mathbb{R}}e^{it\sqrt{(2l+1)|\zeta|}}\chi\left(\frac{2l+1}{|\zeta|},q\right)|\zeta|d\zeta~{}.

We split the integral into ζ>0\zeta>0 and ζ<0\zeta<0. Looking at the first integral and using the new variable s=ζ/2l+1s=\sqrt{\zeta/2l+1}, we get

Il(t)=2(2l+1)20eit(2l+1)sχ(1s2,q)s3𝑑sI_{l}(t)=2(2l+1)^{2}\int_{0}^{\infty}e^{it(2l+1)s}\chi\left(\frac{1}{s^{2}},q\right)s^{3}ds~{}

Observing that the function χ(1s2,q)s3\chi\left(\frac{1}{s^{2}},q\right)s^{3} is a smooth classical symbol supported away of 0, integrations by part give that Il(t)=O(l)I_{l}(t)=O\left(l^{-\infty}\right) as well as its derivatives.

Then we introduce

P¯:=12π02πeitΩPeitΩ𝑑t\bar{P}:=\frac{1}{2\pi}\int_{0}^{2\pi}e^{it\Omega}Pe^{-it\Omega}dt

which is again a pseudo-differential operator. We check that PP¯=[Q,Ω]P-\bar{P}=[Q,\Omega] for a pseudo-differential operator Q: if

St:=0teisΩ(PP¯)eisΩ𝑑t,S_{t}:=\int_{0}^{t}e^{is\Omega}(P-\bar{P})e^{-is\Omega}dt,

this holds true with

Q=12πi02πeitΩSteitΩ𝑑t.Q=\frac{1}{2\pi i}\int_{0}^{2\pi}e^{-it\Omega}S_{t}e^{it\Omega}dt~{}.

The corresponding part of the trace vanishes, because Ω\Omega commutes with Δ3\Delta_{3}. Then the full symbol of P¯\bar{P} can be written as

j=0|ζ|jpj(Iζ,q)\sum_{j=0}^{\infty}|\zeta|^{-j}p_{j}\left(\frac{I}{\zeta},q\right)

with pjp_{j} compactly supported. This allows to reduce to the first case. \square

5 The trace formula for compact quotients of H3H_{3}

This section can be skipped. It contains an example with a direct derivation of Melrose’s trace formula.

Let us give a co-compact subgroup Γ\Gamma of H3H_{3}. We will prove Melrose’s trace formula for M:=Γ\H3M:=\Gamma\backslash H_{3} with Laplacian ΔM\Delta_{M} which is Δ3\Delta_{3}, defined in Section 2.2, restricted to Γ\Gamma-periodic functions. We fix some time T>0T>0 and will look at the trace formula for |t|T|t|\leq T. We choose χD\chi_{D} a smoothed fundamental domain, ie χDC0(H3)\chi_{D}\in C_{0}^{\infty}(H_{3}) with γΓχD(γq)=1\sum_{\gamma\in\Gamma}\chi_{D}(\gamma q)=1. We will denote by D\int_{D}\cdots the integral H3χD\int_{H_{3}}\chi_{D}\cdots.

We start with

eM(t,q,q)=γΓe3(t,q,γq)e_{M}(t,q,q^{\prime})=\sum_{\gamma\in\Gamma}e_{3}(t,q,\gamma q^{\prime})

where e3e_{3} is the half-wave kernel in H3H_{3}. Because of the finite speed of propagation and the fact that Γ\Gamma is discrete, we have only to consider a finite sum for the trace:

Trace(eitΔM)=De3(t,q,q)|dq|+γΓId,minqd(q,γq)TDe3(t,q,γq)|dq|{\rm Trace}\left(e^{it\sqrt{\Delta_{M}}}\right)=\int_{D}e_{3}(t,q,q)|dq|+\sum_{\gamma\in\Gamma\setminus{\rm Id},~{}\min_{q}d(q,\gamma q)\leq T}\int_{D}e_{3}(t,q,\gamma q)|dq|

The first term is smooth outside t=0t=0 while the second one is given by the CDG trace formula.

With more details: let c>0c>0 be small enough so that the cone Cc={g<cζ2}C_{c}=\{g^{\star}<c\zeta^{2}\} has the property (𝒫)(\cal P) that there is no γΓId\gamma\in\Gamma\setminus{\rm Id} with a geodesic from qq to γq\gamma q of length smaller than TT starting with Cauchy data in this cone. This is possible thanks to Lemma 2.1. We can hence split the integrals Iγ(t)=De3(t,q,γq)|dq|I_{\gamma}(t)=\int_{D}e_{3}(t,q,\gamma q)|dq| into two pieces

Iγ(t)=Trace((eitΔ3)τγχDP)+Trace((eitΔ3)τγχD(IdP))I_{\gamma}(t)={\rm Trace}\left(\left(e^{it\sqrt{\Delta_{3}}}\right)\tau_{\gamma}\chi_{D}P\right)+{\rm Trace}\left(\left(e^{it\sqrt{\Delta_{3}}}\right)\tau_{\gamma}\chi_{D}({\rm Id}-P)\right)

with τγ(f)=fγ1\tau_{\gamma}(f)=f\circ\gamma^{-1} and P=ψ(Δ3/|Z|2)P=\psi(\Delta_{3}/|Z|^{2}) where ψ\psi belongs to Co()C_{o}^{\infty}({\mathbb{R}}), is equal to 11 near 0 and is supported in ]c,c[]-{c},{c}[. The first term is smooth by Theorem 3.2 and property (𝒫)(\cal P). The second term corresponds to the elliptic region and hence we use the parametrix for the wave equation given by “FIO”s as given in the CDG trace formula. We get then that the singularities of the wave trace locate on the length spectrum.

Note that the ”heat trace” can be computed from the explicit expression of the spectrum. This is worked out in Appendix E.

6 Normal forms

In what follows, MM is a closed 3D sR manifold of contact type equipped with a smooth volume. We denote by Δ\Delta the associated Laplacian. The proof of the Melrose formula will be done by using a normal form allowing a reduction to the case of Heisenberg.

6.1 Classical normal form

Theorem 6.1

Let Σ\Sigma be the characteristic manifold, i.e. the orthogonal of the distribution with respect to the duality, and let σ0Σ0\sigma_{0}\in\Sigma\setminus 0, then there exists a conical neighbourhood  UU of σ0\sigma_{0} and an homogeneous symplectic diffeomorphism χ\chi of UU onto a conical neighbourhood of (0,0,0;0,0,1)(0,0,0;0,0,1) in TH3T^{\star}H_{3}, so that gH3χ=gMg^{\star}_{H_{3}}\circ\chi=g^{\star}_{M}.

We use first [Me-84] (Prop. 2.3) (or [CHT-21] (Theorem 2.1)) to reduce to ρI\rho I where ρ\rho is the Reeb Hamiltonian and II the harmonic oscillator Hamiltonian: this means that there is an homogeneous canononical transformation χ\chi from a conic neighborhood of σ0\sigma_{0} into Σσ×u,v2\Sigma_{\sigma}\times{\mathbb{R}}^{2}_{u,v} so that ρIχ=g\rho I\circ\chi=g^{\star} with I=u2+v2I=u^{2}+v^{2}. Let us denote by ζ\zeta the principal symbol of ZZ in Equation (1). Then we use the normal form of Duistermaat-Hörmander [DH-72] (Prop. 6.1.3) to reduce ρ\rho to |ζ||\zeta| by a canonical transformation. We get then the normal form |ζ|I|\zeta|I which is the canonical decomposition of gH3g^{\star}_{H_{3}} used in [CHT-18] (see the principal symbols in Equation (1)).

6.2 Quantum normal form

This is a 3-step reduction working in some conical neighbourhood CC of a point of Σ\Sigma.

  1. 1.

    Using FIO’s associated to χ\chi, we first reduce the Laplacian to a pseudo-differential operator  of the form |Z|Ω+R0|Z|\Omega+R_{0} where R0R_{0} is a pseudo-differential operator  of degree 0. This step is worked out in Theorem 5.2 of [CHT-18].

  2. 2.

    We can improve the previous normal form so that R0R_{0} commutes with Ω\Omega: the cohomological equations

    {|ζ|I,a}=b\{|\zeta|I,a\}=b

    (on TH3T^{\star}H_{3}) where bb, vanishing on Σ\Sigma and homogeneous of degree jj can be solved as shown in Appendix D which is an improvement of what is proved in [CHT-18]. It follows that we get a normal form Δ3+R0\Delta_{3}+R_{0} with R0R_{0} commuting with Ω\Omega: the full symbol of R0R_{0} is independent of the angular part of the (u,v)(u,v) variables.

  3. 3.

    Using the spectral decomposition of Ω\Omega, we get a decomposition

    Δl=0(2l+1)ΔlΠl\Delta\equiv\oplus_{l=0}^{\infty}(2l+1)\Delta_{l}\Pi_{l}

    where the Δl\Delta_{l}’s are pseudo-differential operators of the form

    Δl=(|Z|+12l+1R0)\Delta_{l}=\left(|Z|+\frac{1}{2l+1}R_{0}\right)

    and Πl\Pi_{l} is the projector on the eigenspace of eigenvalue 2l+12l+1 of Ω\Omega. We can then use a reduction of the pseudo-differential operators |Z|+12l+1R0|Z|+\frac{1}{2l+1}R_{0} to |Z||Z| by conjugating by elliptic pseudo-differential operators   AlA_{l} depending smoothly of ε=1/(2l+1)\varepsilon=1/(2l+1) and commuting with Ω\Omega as in [DH-72], proposition 6.1.4. We have

    Al1(|Z|+12l+1R0)Al|Z|A_{l}^{-1}\left(|Z|+\frac{1}{2l+1}R_{0}\right)A_{l}\equiv|Z|

    where \equiv means modulo smoothing operators in CC.

7 Proof of the Melrose trace formula

Let us fix T>0T>0 and try to prove Melrose’s trace formula for times tJ:=[T,T]t\in J:=[-T,T]. Let us fix, for each σΣ\sigma\in\Sigma, a conical neighbourhood   UσU_{\sigma} of σ\sigma as in section 6.2. We then take WσVσUσW_{\sigma}\Subset V_{\sigma}\Subset U_{\sigma} so that, for any zVσz\in V_{\sigma} and any tJt\in J, Gt(z)UσG_{t}(z)\in U_{\sigma} where GtG_{t} is the geodesic flow. This is clearly possible using the classical normal form and the Lemma 2.1. We then take a finite cover of Σ\Sigma by open cones Wα:=WσαW_{\alpha}:=W_{\sigma_{\alpha}} and a finite pseudo-differential partition of unity (χ0,χα(αB))(\chi_{0},\chi_{\alpha}(\alpha\in B)) so that WF(χ0)Σ=WF^{\prime}(\chi_{0})\cap\Sigma=\emptyset, χα=Id\chi_{\alpha}={\rm Id} in WαW_{\alpha} and WF(χα)VαWF^{\prime}(\chi_{\alpha})\subset V_{\alpha}. We have then to compute the traces of (costΔ)χ0\left(\cos t\sqrt{\Delta}\right)\chi_{0} and (costΔ)χα\left(\cos t\sqrt{\Delta}\right)\chi_{\alpha}. We will prefer to use the wave equation now because the operator Δ\sqrt{\Delta} is not a pseudo-differential operator! We know from Theorem 3.2 that, for tJt\in J, WF(cos(tΔ)χα)WF^{\prime}\left(\cos(t\sqrt{\Delta})\chi_{\alpha}\right) is a subset of

{(z,z,t,0)|zVα}{(z,G±t(z),t,τ=±g(z))|zVα}.\{(z,z,t,0)|z\in V_{\alpha}\}\cup\{(z,G_{\pm t}(z),t,\tau=\pm g^{\star}(z))|z\in V_{\alpha}\}.

If u(t)=cos(tΔ)χαu0u(t)=\cos(t\sqrt{\Delta})\chi_{\alpha}u_{0}, we have utt+Δu=0,u(0)=χαu0,ut(0)=0.u_{tt}+\Delta u=0,u(0)=\chi_{\alpha}u_{0},u_{t}(0)=0. We can hence use the normal form and denote by \equiv the equality “modulo smooth functions of tJt\in J” to get

Zα(t):=Trace(cos(tΔ)χα)Trace(cos(tΔ3+R0)χα~),Z_{\alpha}(t):={~\rm Trace}(\cos(t\sqrt{\Delta})\chi_{\alpha})\equiv{~\rm Trace}(\cos(t\sqrt{\Delta_{3}+R_{0}})\widetilde{\chi_{\alpha}}),

where χα~\widetilde{\chi_{\alpha}} is the PDO obtained by Egorov theorem when we take the normal form. Then, because R0R_{0} commutes with Ω\Omega,

Zα(t)l=0Trace(cos(t(2l+1)(|Z|+12l+1R0))Πlχα~)Z_{\alpha}(t)\equiv\sum_{l=0}^{\infty}{~\rm Trace}\left(\cos\left(t\sqrt{(2l+1)\left(|Z|+\frac{1}{2l+1}R_{0}\right)}\right)\Pi_{l}\widetilde{\chi_{\alpha}}\right)

and

Zα(t)l=0Trace(Al1cos(t(2l+1)|Z|)AlΠlχα~)Z_{\alpha}(t)\equiv\sum_{l=0}^{\infty}{~\rm Trace}\left(A_{l}^{-1}\cos\left(t\sqrt{(2l+1)|Z|}\right)A_{l}\Pi_{l}\widetilde{\chi_{\alpha}}\right)
Zα(t)l=0Trace(cos(t(2l+1)|Z|)AlΠlχα~Al1).Z_{\alpha}(t)\equiv\sum_{l=0}^{\infty}{~\rm Trace}\left(\cos\left(t\sqrt{(2l+1)|Z|}\right)A_{l}\Pi_{l}\widetilde{\chi_{\alpha}}A_{l}^{-1}\right).

We can assume that AlA_{l} is invertible on WF(χα)WF^{\prime}(\chi_{\alpha}) and put χαl~~=Alχα~Al1\widetilde{\widetilde{\chi_{\alpha}^{l}}}=A_{l}{\widetilde{\chi_{\alpha}}}A_{l}^{-1}. We get

Zα(t)l=0Trace(cos(t(2l+1)|Z|)AlΠlAl1χαl~~)Z_{\alpha}(t)\equiv\sum_{l=0}^{\infty}{~\rm Trace}\left(\cos\left(t\sqrt{(2l+1)|Z|}\right)A_{l}\Pi_{l}A_{l}^{-1}\widetilde{\widetilde{\chi_{\alpha}^{l}}}\right)

Using the fact AlA_{l} commutes with Ω\Omega and hence with Πl\Pi_{l}, we get finally

Zα(t)l=0Trace(cos(t(2l+1)|Z|)Πlχαl~~Πl)Z_{\alpha}(t)\equiv\sum_{l=0}^{\infty}{~\rm Trace}\left(\cos\left(t\sqrt{(2l+1)|Z|}\right)\Pi_{l}\widetilde{\widetilde{\chi_{\alpha}^{l}}}\Pi_{l}\right)

We can then apply a variant of the Proposition 4.1, more precisely of its proof, where PP is replaced by l=0Πlχαl~~Πl\oplus_{l=0}^{\infty}\Pi_{l}\widetilde{\widetilde{\chi_{\alpha}^{l}}}\Pi_{l} and using the fact that the AlA_{l}’s and hence the χαl~~\widetilde{\widetilde{\chi_{\alpha}^{l}}} too are uniformly bounded pseudo-differential operators.

It remains to study the part Z0(t)=Trace(cos(tΔ)χ0)Z_{0}(t)={\rm Trace}\left(\cos(t\sqrt{\Delta})\chi_{0}\right) which involves the elliptic part of the dynamics for which we can use the FIO parametrix as in [DG-75]. More precisely, for tJt\in J, the geodesic flow maps the microsupport of χ0\chi_{0} away of Σ\Sigma, therefore there exists a parametrix for U(t)χ0U(t)\chi_{0} given by Fourier integral operators as in the paper [DG-75] and the calculation of the trace therefore follows the same path.

Appendices

Appendix A Wave and half-wave equations

Let Δ\Delta be a self-adjoint positive sub-elliptic operator on a closed manifold. The wave equation is

2ut2+Δu=0,u(t=0)=u0,ut(t=0)=v0\frac{\partial^{2}u}{\partial t^{2}}+\Delta u=0,u(t=0)=u_{0},\frac{\partial u}{\partial t}(t=0)=v_{0}

This gives a one parameter group U(t)=(U0(t),U1(t))U(t)=(U_{0}(t),U_{1}(t)) on L2×L2L^{2}\times L^{2}. The trace of U0(t)U_{0}(t) is

Z0(t)=Trace(costΔ)=j=1costλjZ_{0}(t)={\rm Trace}\left(\cos t\sqrt{\Delta}\right)=\sum_{j=1}^{\infty}\cos t\sqrt{\lambda_{j}}

One can introduce also the half-wave equation ut=iΔu,u(t=0)=u0\frac{\partial u}{\partial t}=i\sqrt{\Delta}u,~{}u(t=0)=u_{0}. The trace of the half-wave group is Z(t)=j=1eitλjZ(t)=\sum_{j=1}^{\infty}e^{it\sqrt{\lambda_{j}}}. We have the relation Z=2H(Z0)Z=2H(Z_{0}) where HH is the L2L^{2}-projector multiplying the Fourier transform by the Heaviside function. It follows that the singularities of both distributions are easily related.

In the elliptic case, one can work directly with the half-wave group because Δ\sqrt{\Delta} is still an elliptic pseudo-differential operator (Seeley’s Theorem [See-67]). This is no longer the case for sub-elliptic operators.

Appendix B The value of Kζl(m,m)K^{l}_{\zeta}(m,m)

Recall that KζlK_{\zeta}^{l} is the orthogonal projector on the ll-th Landau level with a magnetic field in 2{\mathbb{R}}^{2} equal to ζdxdy\zeta dx\wedge dy. An easy rescaling shows that Kζl(m,m)=|ζ|K1l(m,m)K_{\zeta}^{l}(m,m)=|\zeta|K_{1}^{l}(m,m). We know from the Mehler formula (see [Si-79], p. 168) that the heat kernel of the magnetic Schrödinger operator with constant magnetic field equal to 11 is given on the diagonal by

e(t,m,m)=14πsinhte(t,m,m)=\frac{1}{4\pi\sinh t}

On the other hand, we have

e(t,m,m)=l=0e(2l+1)tK1l(m,m)e(t,m,m)=\sum_{l=0}^{\infty}e^{-(2l+1)t}K_{1}^{l}(m,m)

and

14πsinht=12πl=0e(2l+1)t\frac{1}{4\pi\sinh t}=\frac{1}{2\pi}\sum_{l=0}^{\infty}e^{-(2l+1)t}

Identifying both sums as Taylor series in x=etx=e^{-t} gives

K1l(m,m)=12π.K_{1}^{l}(m,m)=\frac{1}{2\pi}.

Appendix C Toeplitz operators

Let Σ\Sigma be a symplectic cone with a compact basis. Louis Boutet de Montvel and Victor Guillemin associate in [Bo-80, BG-81] to such a cone an Hilbert space and an algebra of operators called the Toeplitz operators with the same properties as the classical pseudo-differential operators. The latter case corresponds to the cone which is a cotangent cone. For an introduction, one can look at [CdV-94].

Two examples are implicitly present in this paper:

  1. 1.

    Harmonic oscillator: the harmonic oscillator Ω=dx2+x2\Omega=-d_{x}^{2}+x^{2} is an elliptic self-adjoint Toeplitz operator; the cone Σ\Sigma is u,v20{\mathbb{R}}^{2}_{u,v}\setminus 0 with the symplectic form dudvdu\wedge dv and the dilations λ.(u,v)=(λu,λv)\lambda.(u,v)=(\sqrt{\lambda}u,\sqrt{\lambda}v). The symbol of Ω\Omega is u2+v2u^{2}+v^{2}.

  2. 2.

    Quantization of the Reeb flow: if ΣTX0\Sigma\subset T^{\star}X\setminus 0 is the characteristic cone of our sR Laplacian, one can quantize the Reeb Hamiltonian ρ\rho as a first order elliptic Toeplitz operator of degree 11.

Appendix D A cohomological equation

The following proposition is a global formulation of the formal cohomological equations discussed in [CHT-18] (section 5.1 and Appendix C) with a simple proof:

Proposition D.1

We consider the cohomological equation

{|ζ|I,A}=B\{|\zeta|I,A\}=B (2)

where A,BA,B are smooth homogeneous functions in the cone C:={I<c|ζ|}C:=\{I<c|\zeta|\} with compact support in qH3q\in H_{3}. If BB is homogeneous of degree jj and vanishes on Σ:={I=0}\Sigma:=\{I=0\}, Equation (2) admits a solution AA homogeneous of degree j1j-1.

Restricting to ζ=1\zeta=1, reduces to prove the following Lemma.

Lemma D.1

Let us consider the differential equation

aθ+12Iaz=b(z,w)\frac{\partial a}{\partial\theta}+\frac{1}{2}I\frac{\partial a}{\partial z}=b(z,w) (3)

with (z,w)×{|w|<c},w=|w|eiθ(z,w)\in{\mathbb{R}}\times\{|w|<c\},w=|w|e^{i\theta}, bb smooth, compactly supported in zz and I=|w|2I=|w|^{2}. We assume that b(z,0)=0b(z,0)=0.

Then Equation (3) admits a smooth solution aa depending smoothly of bb.

Any smooth function ff in some disk in {\mathbb{C}} admits a Fourier expansion

f(w)=n=0fn(|w|2)wn+n=1gn(|w|2)w¯nf(w)=\sum_{n=0}^{\infty}f_{n}(|w|^{2})w^{n}+\sum_{n=1}^{\infty}g_{n}(|w|^{2})\bar{w}^{n}

where the fnf_{n}’s and the gng_{n}’s are smooth 333This is proved first in formal power series in (w,w¯)(w,\bar{w}). This reduces the problem to functions which are flat at the origin. A Fourier expansion in θ\theta allows to finish the proof.. We can use this expansion with f=a(z,.)f=a(z,.) and g=b(z,.)g=b(z,.). We consider only the sum of powers of wnw^{n}. The second can be worked out in a similar way. We then put a=n=1an(z,I)wn+a0(z,I)a=\sum_{n=1}^{\infty}a_{n}(z,I)w^{n}+a_{0}(z,I) and b=n=1bn(z,I)wn+Ic0(z,I)b=\sum_{n=1}^{\infty}b_{n}(z,I)w^{n}+Ic_{0}(z,I). The factorization b0(z,I)=Ic0(z,I)b_{0}(z,I)=Ic_{0}(z,I) follows from the assumption on bb. We can take a0(z,I)=2zb0(s,I)𝑑sa_{0}(z,I)=2\int_{-\infty}^{z}b_{0}(s,I)ds. The equation for ana_{n}, with n1n\geq 1, writes

inan+12Ianz=bnina_{n}+\frac{1}{2}I\frac{\partial a_{n}}{\partial z}=b_{n}

We can solve it, for I0I\neq 0, by

an(z,I)=2I0bn(z+u,I)ein2u/I𝑑u=0bn(z+Is/2,I)eins𝑑sa_{n}(z,I)=\frac{2}{I}\int_{-\infty}^{0}b_{n}(z+u,I)e^{in2u/I}du=\int_{-\infty}^{0}b_{n}(z+Is/2,I)e^{ins}ds

and an(z,0)=bn(z,0)/ina_{n}(z,0)=b_{n}(z,0)/in. We need to prove that ana_{n} is smooth: we check that ana_{n} is continuous and then the derivative are given by the same kind of integrals with derivatives of bnb_{n} as shown by the last expression of ana_{n}. The continuity of ana_{n} follows from an integrations by parts in the first expression of ana_{n}.

Now we want to add up the series nanwn\sum_{n}a_{n}w^{n}. I was not able to do that directly and will proceed as follows: the sum nanwn\sum_{n}a_{n}w^{n} is convergent as a formal series along Σ\Sigma, because anwn=O(In/2)a_{n}w^{n}=O(I^{n/2}). Using Borel procedure, we need only to solve our cohomological equation with a flat righthandside. This follows clearly from the expression

a(z,w)=0b(z+It/2,eitw)𝑑t.a(z,w)=\int_{-\infty}^{0}b(z+It/2,e^{it}w)dt~{}.

In fact only the behaviour as I0I\rightarrow 0 could be a problem, but we can divide bb by any power of II.

Appendix E The heat trace for a compact quotient of H3H_{3}

Let us consider the discrete subgroup Γ=(2π)2×π\Gamma=\left(\sqrt{2\pi}{\mathbb{Z}}\right)^{2}\times\pi{\mathbb{Z}} of H3H_{3} identified with x,y,z3{\mathbb{R}}^{3}_{x,y,z} as in Section 2.2. The spectrum of the sub-Laplacian defined in section 2.2 on M=H3/ΓM=H_{3}/\Gamma is the union of the spectrum of the flat torus 2/(2π)2{\mathbb{R}}^{2}/\left(\sqrt{2\pi}{\mathbb{Z}}\right)^{2} and the eigenvalues 2m(2l+1),m1,l0,2m(2l+1),m\geq 1,~{}l\geq 0, with multiplicities 2m2m. The corresponding part of the complexified heat trace is hence

Zo(z)=m=12ml=0e2m(2l+1)zZ_{o}(z)=\sum_{m=1}^{\infty}2m\sum_{l=0}^{\infty}e^{-2m(2l+1)z}

with (z)>0\Re(z)>0. Summing with respect to ll gives

Zo(z)=m=1msinh2mzZ_{o}(z)=\sum_{m=1}^{\infty}\frac{m}{\sinh 2mz}

which we rewrite as

Zo(z)=14zm2mzsinh2mz14zZ_{o}(z)=\frac{1}{4z}\sum_{m\in{\mathbb{Z}}}^{\infty}\frac{2mz}{\sinh 2mz}-\frac{1}{4z}

The Fourier transform of xsinhx\frac{x}{\sinh x} is π21+coshπξ\frac{\pi^{2}}{1+\cosh\pi\xi}. Applying Poisson summation formula to the last expression of ZoZ_{o}, we get

Zo(z)=π216z214z+π24z2n=111+coshπ2n/zZ_{o}(z)=\frac{\pi^{2}}{16z^{2}}-\frac{1}{4z}+\frac{\pi^{2}}{4z^{2}}\sum_{n=1}^{\infty}\frac{1}{1+\cosh\pi^{2}n/z}

The first term gives the Weyl law. Each term in the sum w.r. to nn is equivalent to

π22z2eπ2n/z\frac{\pi^{2}}{2z^{2}}e^{-\pi^{2}n/z}

We observe that the lengths of the periodic geodesic of MM are the numbers 2πn2\pi\sqrt{n}. Hence we recover also the length spectrum giving contributions of the order of exp(L2/4z){\rm exp}(-L^{2}/4z) as in the Riemannian case as proved in [CdV-73]. It could be nice to derive an exact formula for the wave trace from our expression of the heat trace. Similarly, the heat trace for the Riemannian Laplacian on MM was computed by Hubert Pesce [Pe-94].

Appendix F A short history of the trace formulae

The trace formulae were first discovered independently by two groups of physicists: Martin Gutzwiller [Gu-71] for a semi-classical Schrödinger operator and Roger Balian & Claude Bloch in a very impressive series of papers for Laplacians in Euclidean domains [Ba-Bl-70, Ba-Bl-71, Ba-Bl-72]. In [Ba-Bl-72], page 154, the authors suggested already a possible application to the inverse spectral problems444They wrote “The analysis of the eigenvalue density as a sum of oscillating terms gives a new insight into the problem of “hearing the shape of a drum” [Kac paper]. …It is convenient, for the discussion to start from the fact that the knowledge of eigenvalues determines uniquely the path generating function …Thus …the lengths of the closed stationary polygons are determined”. , an industry which just started at the end of the sixties. From the point of view of mathematics, the Poisson summation formula can be interpreted as a trace formula for the Euclidian Laplacian on flat tori. Similarly, the famous Selberg trace formula [Se-56] (see also Heinz Huber [Hu-59]) is a trace formula for the Laplacian on hyperbolic surfaces. Then my thesis [CdV-73], inspired by the work of Balian and Bloch and the Selberg trace formula, uses the complex heat equation for general closed Riemannian manifold. The definitive version, the CDG formula, using wave equation, was discovered by Jacques Chazarain and the tandem Hans Duistermaat & Victor Guillemin in [Ch-74, DG-75]. They use the power of the Fourier Integral Operators calculus [Ho-71, DH-72]. See [CdV-07] for a review paper. Later results cover the cases of manifolds with boundaries and semi-classical versions.

Author references: no conflist of interest, no funding.

References

  • [Ba-Bl-70] Roger Balian & Claude Bloch. Distribution of eigenfrequencies for the wave equation in a finite domain I. Ann. of Physics 60:401 (1970).
  • [Ba-Bl-71] Roger Balian & Claude Bloch. Distribution of eigenfrequencies for the wave equation in a finite domain II. Ann. of Physics 64:271 (1971).
  • [Ba-Bl-72] Roger Balian & Claude Bloch. Distribution of eigenfrequencies for the wave equation in a finite domain III. Ann. of Physics, 69:76 (1972).
  • [Bo-80] Louis Boutet de Monvel. Opérateurs à coefficients polynomiaux, espace de Bargmann, et opérateurs de Toeplitz. Séminaire Équations aux dérivées partielles (Polytechnique) exposé no 2 bis. (1980-1981).
  • [BG-81] Louis Boutet de Monvel & Victor Guillemin. The spectral theory of Toeplitz operators. Ann. Math. Studies 99 (1981).
  • [CdV-73] Yves Colin de Verdière. Spectre du Laplacien et longueurs des géodésiques périodiques II. Compositio Mathematica 27(2):159-184 (1973).
  • [CdV-94] Yves Colin de Verdière. Une introduction aux opérateurs de Toeplitz. Séminaire de Théorie spectrale et géométrie, 13:135–141 (1994-1995).
  • [CdV-07] Yves Colin de Verdière. Spectrum of the Laplace operator and periodic geodesics, thirty years after. Ann. Inst. Fourier  57:2429– 2463 (2007).
  • [CdV-22] Yves Colin de Verdière. Periodic geodesics for contact sub-Riemannian 3D manifolds. ArXiv 2202.13743 (2022).
  • [Ch-74] Jacques Chazarain. Formule de Poisson pour les variétés riemanniennes. Inventiones Math. 24:65–82 (1974).
  • [CHT-18] Yves Colin de Verdière, Luc Hillairet & Emmanuel Trélat. Spectral asymptotics for sub-Riemannian Laplacians I: Quantum ergodicity and quantum limits in the 3D contact case, Duke Math. Journal 167 (1) (2018).
  • [CHT-21] Yves Colin de Verdière, Luc Hillairet & Emmanuel Trélat. Spiraling of sub-Riemannian geodesics around the Reeb flow in the 3D contact case. ArXiv 2102.12741 (2021).
  • [DG-75] Hans Duistermaat & Victor Guillemin. The spectrum of positive elliptic operators and periodic bicharacteristics. Inventiones Math. 29:39–79 (1975).
  • [DH-72] Hans Duistermaat & Lars Hörmander. Fourier Integral Operators II. Acta Mathematica 128:183–269 (1972).
  • [Gu-71] Martin Gutzwiller.Periodic orbits and classical quantization conditions. J. Math. Phys. 12:343–358 (1971).
  • [Hej-76] Dennis Hejhal. The Selberg trace formula and the Riemann ζ\zeta function. Duke Math. J. 43: 441–482 (1976).
  • [Ho-71] Lars Hörmander. Fourier Integral Operators I. Acta Mathematica 127:79–183 (1971).
  • [Hu-59] Heinz Huber, Zur analytischen Theorie hyperbolischen Raumformen und Bewegungsgruppen. Math. Ann., 138:1–26 (1959).
  • [Iv-76] Victor Ivrii. Wave front of some hyperbolic equations. Sov. Math. Dokl. 226(5):1009–1011, 229(2):280–283 (1976).
  • [La-82] Bernard Lascar. Propagation des singularités pour des équations hyperboliques à caractéristiques de multiplicité au plus double et singularités Masloviennes. American J. Maths 104:227–285 (1982).
  • [Le-20] Cyril Letrouit. Subelliptic wave equations are never observable. An. & PDE (to appear), arXiv:2002.01259 (2020).
  • [Le-21] Cyril Letrouit. Propagation of singularities for subelliptic wave equations. ArXiv:2106.07216 (2021).
  • [Me-84] Richard B. Melrose. The wave equation for a hypoelliptic operator with symplectic characteristics of codimension two. J Analyse math. 44:134–182 (1984-1985).
  • [Me-86] Richard B. Melrose. Propagation for the wave group of a positive subelliptic second-order differential operator. In : Hyperbolic equations and related topics. Academic Press, p. 181–192 (1986).
  • [Pe-94] Hubert Pesce. Une formule de Poisson pour les variétés de Heisenberg. Duke Math. J. 73:1–17 (1994).
  • [See-67] Robert Seeley. Complex powers of an elliptic operator. Proc. Sympos. Pure Math. 10:288–307 (1967).
  • [Se-56] Atle Selberg. Harmonic analysis and discontinuous groups in weakly symmetric Riemannian spaces with applications to Dirichlet series. J. Indian Math. Soc. 20: 47–87 (1956).
  • [Si-79] Barry Simon. Functional Integration and Quantum Physics. Academic Press (1979).