On the geometry of $Diff(S^1)-$pseudodifferential operators based on renormalized traces

In this article, we examine the geometry of a group of Fourier-integral operators, which is the central extension of $Diff(S^1)$ with a group of classical pseudo-differential operators of any order. Several subgroups are considered, and the corresponding groups with formal pseudodifferential operators are defined. We investigate the relationship of this group with the restricted general linear group $GL_{res},$ we define a right-invariant pseudo-Riemannian metric on it that extends the Hilbert-Schmidt Riemannian metric by the use of renormalized traces of pseudo-differential operators, and we describe classes of remarkable connections.


Introduction
In the mathematical literature, the introduction of infnite dimensional Lie groups is very often motivated by the use that can be done of such objects which intrinsic properties are difficult to catch. For example, symmetry groups arise in the theory of ordinary differential equations and partial differential equations. Groups of diffeomorphisms arise in the basic theory of differential manifolds, dynamical systems, and have applications in a wide range of examples such as knot theory, stochastic analysis. Manifolds of maps, current groups and gauge groups have their own applications in various models in physics while their intrinsic properties are deeply related to homotopy theory.
The aim of this paper is the description of a family of infinite dimensional Lie groups, derived from a natural interplay between classical pseudodifferential operators and diffeomorphisms, which can be found in the litterature first in [48] and in [35], specializing the study when the base manifold is S 1 . Motivations for the introduction of these groups are different in the two works. In [48] the motivation is the full description of a possible structure group for infinite dimensional principal bundles where a Chern-Weil theory can be stated. In [35], the motivation comes from the need of description of the structure group of the space of non-parametrized space of embeddings of a smooth compact boundaryless manifold M to a smooth finite dimensional Riemanian manifold manifold N. In both descriptions, the group under consideration is the central extension of a group of diffeomorphisms by a group of classical pseudo-differential operators.
More precisely, the group under consideration in a central extension of the group of diffeomoprhisms Dif f (S 1 ) by the group Cl * (S 1 , V ) of invertible elements of the algebra Cl(S 1 , V ) of non-formal, classical, maybe unbounded pseudo-differential operators acting on a trivial n−dimensional Hermitian bundle S 1 × V over S 1 . In such an object, one can derive many infinite dimensional groups: (1) the loop group C ∞ (S 1 , SU n ) (2) the group of orientation preserving diffeomorphisms Dif f + (S 1 ) (3) the group generated by L 2 −orthogonal symmetries related to L 2 orthogonal projections on a finite demensional vector subspace of C ∞ (S 1 , V ) But beside these examples of groups of bounded operators, which are all subroups of the restricted unitary group U res described in [49], we also wish to recover here some non-formal versions of the spaces of elliptic injective classical pseudo differential operators, and in particular the square root of the Laplacian, in order to state results in a framework as general as possible. Due to the presence of unbounded pseudodifferential operators, and in particular differential operators of order 1, the Lie algebra of this group cannot be embedded in a group of bounded operators acting on the space of sections C ∞ (S 1 , V ), but only represented in it. As a technical remark, we have to say that we have here an example of non-regular infinite dimensional Lie group, as given in remark 1.27 adapting a remark from [37]. But another technical feature is that this group does not seem to carry atlas, for the same reasons of presence of unbounded operators, as first described in [1] in the context of formal pseudo-differential operators. But one can anyway consider this group as an "infinite dimensional group" in the way of [18], or as a so-called Frölicher Lie group, see e.g. [37] and references therein, in order to make safe the notion of smoothness. In the section dedicated to the preliminaries, we describe this groups and some of its remarkable subgroups, we recall classical propoerties of the (zeta-))renormalized traces of pseudo-differential operators, describe useful splisstings of the algebra of formal pseudo-differntial operators from existing literature, and fully describe the space of formal Dif f (S 1 )−pseudo-differential operators. This last description is, to our knowledge, not given before this work.
In section 2, we compare this group with the restricted general linear group and develop the index 2 − f orm λ on it. Indeed, by considering only S 1 as a base manifold, this enables us to describe more deeply the algebraic and geometric structurees derived from the Dirac operator D, or more precisely its sign ǫ(D), which defines a polarization on L 2 (S 1 , C k ), splitting this space into eigenspaces of positive and non-positive eigenvalues. This polarization is described in e.g. [49], is shown here to generalize to unbounded operators in an elegant way, generalizing the remarks initiated in [28,30] on one hand and some geometric constructions using the Lie group of bounded operators on the other hand. This Lie group has a central extension, and the corresponding central extension of its Lie algebra is given by the Schwinger cocycle, first found by J. Schwinger in [52]. This 2-cocycle was known as a cocycle on Lie algebras of bounded operators (see e.g. [7], [41]). In [30], [32] we proved that the Schwinger cocycle can be extended naturally to the algebra P DO(S 1 , C k ) of (maybe non classical, maybe unbouded) pseudo-differential operators. Moreover, we gave its relations with a pull-back c D + of the Khesin-Kravchenko-Radul cocycle on formal symbols [21], [50], using an appropriate linear extension of the usual trace of traceclass operators. c D + and 1 2 c D S have the same cohomology class. On loop groups, this (Lie algebra)-cocycle pulls back to the central extension of the loop algebra [49], [7], [28], while it enables to recover the Gelfand-Fuchs cocycle on V ect(S 1 ) [30]. We extend here the Lie-algebraic considerations of [30] to the grup under consideration.
In section 3 we construct the extension of the Hlibert-Schmidt Hermitian product to the Lie algebra Cl(S 1 , V ) ⋊ V ect(S 1 ) by replacing the usual trace of trace-class operators by one of its linear extensions, the zeta-renormalized trace tr ∆ described in the preliminaries. The first difficulty comes from the fact that tr ∆ is not tracial, i.e.
. Surprisingly, we obtain a non-degenerate sesquilinear form (., .) ∆ on Cl(S 1 ; V ), whicch gives rise to a non-degenerate bilinear form Re(., .) ∆ on Cl(S 1 ; V )⋊V ect(S 1 ). We also show that these forms are not positive, and multiplication operators are isotropic vectors. These two forms then generate, by right-invariant action, pseudo-Hermitian and pseudo-Riemannian metrics on Cl * (S 1 , V ) and F CL * Due to the presence of a pseudo-Hermitian metric, one can ask whether there exists pseudo-Hermitian connections. This is the aim of section 4, where families of pseudo-Hermitian connections are described. However, there exist some technical difficulties to pass in order to describe the whole family of pseudo-Hermitian connections for (., .) ∆ , due in particular to the actual lack of knowledge on the linear maps acting on Cl(S 1 , V ). We restrict our investigations to connection 1−forms that read as composition by a smoothing operator. Our results are mostly based on the remark that [a, ǫ(D)] and sas * , when s is smoothing, are smoothing operators.
We finish this work by giving some specialization remarks. We describe how Dif f (S 1 ) acts on the polarization, we give another interpretation of the Schwinger cocycle in terms of the curvature of one of our smoothing connections defined in section 4, we show that a Levi-Civita type exists for Re(., .) ∆ on the group F Cl * ee,Dif f (S 1 ) (S 1 , V ) and that it fits with classical formulas for the Levi-Civita connections in finite dimensional settings, and we finish with a blockwise decomposition of Re(., .) ∆ on the group of bounded, even-even class operators F Cl 0, * ee,Dif f (S 1 ) (S 1 , V ).
Acknowledgements: The author is happy to dedicate this work to Sylvie Paycha on her 60th birthday, 18 years after working under her supervision, with gratitude for this first step into research.

Preliminaries on operators on S 1
In this section, we make a moderate use of the notion of diffeology, notion which is not among the main goals of this paper. A classical presentation on diffeologies can be found in [16], and the necessary material is already reviewed in many publications of the author. We refer to [39] and references therein for an updated review of the necessary notions to understand in-depth technical properties of the present paper, and to [35] for a preliminary work about the same objects. When precise notions and properties will be necessary, precise citations will be given. For a superficial reading, the reader can replace diffeologies by a "natural differentiation" on operators.
1.1. Basics on pseudo-differential operators. An exposition of basic facts on pseudo-differential operators can be found in [13]. In this section, we only review the definitions and tools that are necessary for this note.
We set S 1 = {z ∈ C | |z| = 1}. We shall use for convenience the smooth atlas A of S 1 defined as follows: Associated to this atlas, we fix a smooth partition of the unit {s 0 ; s 1 }. We identify each of these functions with its associated multiplication operator when necessary. An operator A : C ∞ (S 1 , C) → C ∞ (S 1 , C) can be described in terms of 4 operators A scalar pseudo-differential operator of order o is an operator (In these formulas, the maps f , s n and A m,n (f ) are read on the local charts ϕ 0 , ϕ 1 , but we preferred to only mention this aspect and not to give heavier formulas and notations, since the setting for S 1 -pseudo-differential operators is rather more simple than for manifolds of higher dimension.) Let E = S 1 × C k be a trivial smooth vector bundle over S 1 . Let s ∈ R. An operator acting on C ∞ (S 1 , C n ) is a pseudo-differential operator if it can be viewed as a (n × n)-matrix of scalar pseudo-differential operators. A pseudo-differential operator A of order o extends to a linear bounded operator on Sobolev spaces H s (S 1 , C n ) → H s−o (S 1 , C n ). In particular, an order 0 pseudo-differential operator is a bounded operator on H s (S 1 , C n ).
A pseudo-differential operator of order o is called classical if and only if its symbols σ m,n have an asymptotic expansion where the maps (σ m,n ) j : S 1 × R * → C, called partial symbols, are j-positively homogeneous, i.e. ∀t > 0, ( We also define log-polyhomogeneous pseudo-differential operators. A pseudodifferential operator is log-polyhomogeneous if and only if there exists o ′ ∈ N such that its symbols σ m,n have an asymptotic expansion where the maps (σ m,n ) j : S 1 × R * → C are classical partial symbols of order j. o ′ is called the logarithmic order of the pseudo-differential operator. Of course, classical pseudo-differential operators are log-polyhomogeneous pseudo-differential operators of logarithmic order 0.
The notion of symbol and partial symbol appear local (dependent on the charts of the atlas) in view of these definitions but, in this very special case of atlas on S 1 where the changes of coordinates are translations, one can see with the formulas of change of coordinates given in e.g. [13] that the partial symbols of a pseudodifferential operator can be defined globally, taking There is another way to define globally the partial symbols of an operator is in [6] , see e.g. [56], using linearizations of the manifold. This second way to define the formal symbol is more useful when one works with manifolds more complicated than S 1 . Now, define the sets of smooth maps The set S(S 1 , C k )/S −∞ (S 1 , C k ) can be understood as the set of asymptotic expansions when ξ → ±∞ up to rapidly decreasing maps in the ξ variable. Then, we define the following multiplication rule some equivalence classes of maps σ and σ ′ in S(S 1 , C k )/S −∞ (S 1 , C k ): Notations. We note by Cl(S 1 , C k )) classical pseudo-differential operators acting on smooth sections of E, and by Cl o (S 1 , C k ) the space of classical pseudodifferential operators of order o.
There is not an isomorphism between the set of symbols an the set of pseudodifferential operators. If we set we notice that it is a two-sided ideal of Cl(S 1 , C k ), and we define the quotient algebra called the algebra of formal pseudo-differential operators. F Cl(S 1 , C k ) is isomorphic to the set of formal symbols [6], and the identification is a morphism of Calgebras, for the multiplication on formal symbols defined before.

1.2.
Renormalized traces. E is equiped this an Hermitian products < ., . >, which induces the following L 2 -inner product on sections of E: where dx is the Riemannian volume.
Definition 1.1. Q is a weight of order s > 0 on E if and only if Q is a classical, elliptic, self-adjoint, positive pseudo-differential operator acting on smooth sections of E.
Recall that, under these assumptions, the weight Q has a real discrete spectrum, and that all its eigenspaces are finite dimensional. For such a weight Q of order q, one can define the complex powers of Q [51], see e.g. [7] for a fast overview of technicalities. The powers Q −s of the weight Q are defined for Re(s) > 0 using with a contour integral, where Γ is a contour around the real positive axis. Let A be a log-polyhomogeneous pseudo-differential operator. The map ζ(A, Q, s) = s ∈ C → tr (AQ −s ) ∈ C , defined for Re(s) large, extends on C to a meromorphic function [26]. When A is classical, ζ(A, Q, .) has a simple pole at 0 with residue 1 q res W A, where res W is the Wodzicki residue ( [57], see also [17]). Notice that the Wodzicki residue extends the Adler trace [3] on formal symbols. Following textbooks [47,53] for the renormalized trace of classical operators, we define Definition 1.2. Let A be a log-polyhomogeneous pseudo-differential operator. The finite part of ζ(A, Q, s) at s = 0 is called the renormalized trace tr Q A. If A is a classical pseudo-differential operator, If A is trace class acting on L 2 (S 1 , C k ), tr Q (A) = tr(A). The functional tr Q is of course not a trace. In this formula, it appears that the Wodzicki residue res W (A).

Proposition 1.3.
(i) The Wodzicki residue res W is a trace on the algebra of classical pseudo-differential operators In particular, res W does not depend on the choice of Q.
Since tr Q is a linear extension of the classical trace tr of trace-class operators acting on L 2 (S , V ), it has weaker properties. Let us summarize some of them which are of interest for our work following first [7], completed by [35] for the third point.
• Given two (classical) pseudo-differential operators A and B, given a weight Q, • Given a differentiable family A t of pseudo-differential operators, given a differentiable family Q t of weights of constant order q, • Under the previous notations, if C is a classical elliptic injective operator or a diffeomorphism, tr C −1 QC C −1 AC is well-defined and equals tr Q A. • Finally, Since tr Q is not tracial, let us give more details on the renormalized trace of the bracket, following [29]. Definition 1.5. Let E be a vector bundle over S 1 , let Q a weight and let a ∈ Z. We define : When needed and appropriate, other properties of renormalized traces will be given later.
-its kernel E 0 , built of constant maps -E + , the vector space spanned by eigenvectors related to positive eigenvalues -E − , the vector space spanned by eigenvectors related to negative eigenvalues. The L 2 −orthogonal projection on E 0 is a smoothing operator, which has null formal symbol. By the way, concentrating our attention on the formal symbol of operators first, we can ignore this projection and hence we work on E + ⊕ E − . The following elementary result will be useful for the sequel. Let us now give a trivial but very useful lemma: Lemma 1.9.
[28] Let f : R * → V be a 0-positively homogeneous function with values in a topological vector space V . Then, for any n ∈ N * , f (n) = 0 where f (n) denotes the n-th derivative of f .
The splitting with induced by the connected components of T * S 1 −S 1 .. In this section, we define two ideals of the algebra F Cl( . This decomposition is explicit in [17, section 4.4., p. 216], and we give an explicit description here following [28,30]. are clearly smooth algebra morphisms (yet non-unital morphisms) that leave the order invariant and are also projections (since multiplication on formal symbols is expressed in terms of pointwise multiplication of tensors).
Since p + is a projection, we have the splitting Let us give another characterization of p + and p − . Looking more precisely at the formal symbols of p E+ and p E− computed in Lemma 1.7, we observe that In particular, we have that p + and p − satisfy Corollary 1.10. Moreover, their symbol do not depend on x. From this, we have the following result.
The "odd-even" splitting. We note by σ(A)(x, ξ) the total formal symbol of A ∈ F Cl(S 1 , V ). The following proposition is trivial: This map is smooth, and Following [53], one can define even-odd class pseudo-differential operators Remark 1.15. In e.g. [19,20], even-even class pseudodifferential operators are called odd class pseudodifferential operators. By the way, following the terminology of [19,20] even-odd class pseudo-differential operators should be called even class. In this paper we prefer to fit with the terminology given in the textbooks [47,53] Proposition 1.16. φ is a projection and F Cl eo (S 1 , V ) = Imφ.
By the way, we also have We have the following composition rules for the class of a formal operator A • B : A even-even class A even class 1.4. Formal and non-formal Dif f (S 1 )−pseudodifferential operators. It follows from [10,45] that Dif f + (S 1 ) is open in the Fréchet manifold C ∞ (S 1 , S 1 ). This fact makes it a Fréchet manifold and, following [45], a regular Fréchet Lie group. In addition to groups of pseudo-differential operators, we also need a restricted class of groups of Fourier integral operators which we will call Dif f (S 1 )−pseudodifferential operators following [35,38]. These groups appear as central extensions of Dif f (S 1 ) or Dif f + (S 1 ) by groups of pseudo-differential operators. We do not state the basic facts on Fourier integral operators here (they can be found in the classical paper [14]). The pseudo-differential operators considered here can be classical, odd class, or anything else. Applying the formulas of "changes of coordinates" (which can be understood as adjoint actions of diffeomorphisms) of e.g. [13], we obtain that eveneven and even-odd class pseudo-differential operators are stable under the adjoint action of Dif f (S 1 ). Thus, we can define the following groups [35]: V ) and g ∈ Dif f (S 1 ) .
(2) The group F Cl 0, * Dif f (S 1 ) (S 1 , V ) is the infinite dimensional group defined by Remark 1.20. This construction of phase functions of Dif f (S 1 )−pseudo-differential operators differs from the one described by Omori [45] and Adams, Ratiu and Schmid [2] for some groups of Fourier integral operators; the exact relation among these constructions still needs to be investigated.
Remark 1.21. The decomposition A = B • g is unique [35], and the diffeomorphism appears as the phase of the Fourier integral operator. Remark 1.23. We have, on these Lie groups, somme difficulties to exhibit an atlas especially when considering unbounded operators. One then can consider "natural" notions of smoothness, inherited from the embedding into Cl(S 1 , V ) for the pseudodifferential part, and from the well-known structure of ILB Lie group [45] from the diffeomorphism (phase) component. In order to be more rigorous, one can then consider Frölicher Lie groups along the lines of [35,38] in this context, or in [36,37] when dealing with other examples where this setting is useful. A not-so-complete description of technical properties of Frölicher Lie groups can be found in works by other authors [4,25,42] but this area of knowledge, however, still needs to be further developed. This lack of theoretical knowledge is not a problem for understanding and dealing with groups of Dif f (S 1 )−pseudo-differential operators.
From this last remark, we deduce that the Lie algebra of F Cl * Dif f (S 1 ) (S 1 , V ) is Cl(S 1 , V ) ⊕ V ect(S 1 ), and we remark that (a, X) → a + X is a Lie algebra morphism with values in Cl(S 1 , V ). The same remark holds for subgroups of F Cl * Dif f (S 1 ) (S 1 , V ). Let us now define a relation of equivalence "up to smoothing operators".
The set of equivalence classes with respect to ≡ is noted as F F Cl * Dif f (S 1 ) (S 1 , V ) and is called the set of formal Dif f (S 1 )−pseudodifferential operators.
The same spaces of formal operators can be constructed using orientation-preserving diffeomorphisms of S 1 , even-even class pseudodifferential operators and so on. We do not feel the need to give here redundant constructions, and obvious notations.
Let A ∈ F Cl * Dif f (S 1 ) (S 1 , V ) and let Id + S ∈ G with S ∈ Cl −∞ (S 1 , C). Then We have with B ∈ Cl * (S 1 , V ) and g ∈ Dif f (S 1 ). As an operator acting on L 2 (S 1 , V )functions, composition on the right by a diffeomorphism is a bounded operator. As an operator from C ∞ (S 1 , V ) to C ∞ (S 1 , V ), composition on the right by a diffeomorphism is also a bounded operator. A pseudo-differential operator of order o is a bounded operator from L 2 (S 1 , V ) to H −o (S 1 , V ), and also from C ∞ (S 1 , V ) to C ∞ (S 1 , V ). Finally a smoothing operator is bounded from any Sobolev space H −o to C ∞ . By the way, Hence G ⊳ F Cl * Dif f (S 1 ) (S 1 , V ). Let us now examine the following diagram: (1.5) The three squares commute, the two horizontal lines are short exact sequences as well as the central culumn. By the way, we also have that as a (set theoric) product can be equiped by the product topology of F Cl * (S 1 , V ) × Dif f (S 1 ). This makes of it a smooth Lie group modelles on a locally convex topological vector space, and we can state: , and its Lie algebra (defined by germs of smooth paths) reads as F Cl(S 1 , V ) ⋊ V ect(S 1 ).
Remark 1.27. It is proven in [35] that, in an algebra of formal pseudo-differential operators that can be identified with F Cl ee (S 1 , C) in our context, the constant vector field t → d dx does not integrate to a smooth path on the group, in other words This shows that F Cl * ee (S 1 , C), and hence Cl * (S 1 , V ) and also F Cl * Dif f (S 1 ) (S 1 , V ) are not regular in the sense of Omori [45], while the same constant vector, understood as an element of V ect(S 1 ), the Lie algebra of Dif f (S 1 ), integrates to a rotation on the circle. This shows that one has to be careful on which component the differential monomials of degree 1 are considered while working with F Cl * Dif f (S 1 ) (S 1 , V ). Its Lie algebra cannot be embedded but only represented in Cl ee (S 1 , V ).

Relation with the restricted linear group
2.1. On cocycles on Cl(S 1 , C k ). Let us first precise which polarization we choose on L 2 (S 1 , C k ). We can choose independently two polarizarions : -one setting H Since E 0 is of dimension k, the orthogonal projection on E 0 is a smoothing operator. Hence, σ(p H (1) We introduce the notation, for A ∈ P DO(S 1 , C k ), + , and we set ǫ(D) = p H+ − p H− . We notice that σ(A ++ ) = σ + (A), and recall the following result [30]: is a well-defined 2-cocycle on P DO(S 1 , C k ). Moreover, c D s is non trivial on any Lie Along this cocycle, we have to mention two others. First, the Kravchenko-Khesin cocycle [21], defined on Adler series a = n≤k a n ξ n and b = m≤l b m ξ m by which pulls-back on the algebra Cl(S 1 , C k ), following a procedure first described by Radul [50] to our knowledge, to a cocycle called Kravchenko-Khesin-Radul cocycle in [31], defined in its polarized version by which is the pull-back, up to the constant 1 2π , of c KK through the map A ∈ Cl(S 1 , C k ) → σ + (A).
Secondly, the index cocycle described in [49] and extended to Cl(S 1 , C k ) in [30], defined by: . We have to remark that, in order to have a well-defined formula, the operator [A ++ , B ++ ]−[A, B] ++ needs to be trace-class. When A and B are pseudo-differential operators, this operator is smoothing and hence the trace is well-defined. Following [30], we can state: C k ), the cocycles c KKR , λ and 1 2 c s are non trivial in Hoschild cohomology, and belong to the same cohomology class.

GL res and its subgroups of Fourier-integral operators.
Let us now turn to the Lie group of bounded operators described in [49]: Let us now give a new light on an old result present in [49] from a topological viewpoint, expressed by remarks but not stated clearly in the mathematical litterature to our knowledge. We choose here a new approach for the proof, more easy and much more fast, and adapted to our approach of (maybe generalized) differentiability pior to topological considerations.
Proof. Let us assume that the injection map Dif f + (S 1 ) ֒→ GL res (S 1 , C k ) is differentiable. The group GL res (S 1 , C k ) is acting smoothly on L 2 (S 1 , C n ) and hence the Lie algebra of GL res (S 1 , C k ) is a Lie algebra of bounded operators acting on L 2 (S 1 , C n ). The Lie algebra V ect(S 1 ) is a Lie algebra of unbounded operators acting on L 2 (S 1 , C n ) hence the injection map Dif f + (S 1 ) ֒→ GL res (S 1 , C k ) is not differentiable.
From this Lemma, next theorem is straightforward: Let us now recall a result from [49]: Proposition 2.6. Let us consider the right-invariant 2 form generated by λ on GL res , that we note byλ. Thenλ is a closed, non exact 2-form on GL res .
Lemma 2.7. Let γ : R → F Cl * Dif f+(S 1 ) (S 1 , C k ) be a smooth path. Then the path is a smooth path of smoothing operators with respect to any of these differentiable structures: • the differentiable structure of F Cl * Dif f+(S 1 ) (S 1 , C k ) • the differentiable structure of GL res (S 1 , C k ).
As a consequence, we get: 2 c s generates a closed, non exact 2-form on F Cl * Dif f+(S 1 ) (S 1 , C k ).

Renormalized extension of the Hilbert-Schmidt Hermitian metric
The vector space Cl −1 (S 1 , V ) is a space of Hilbert-Schmidt operators. As a subspace, Cl −1 (S 1 , V ) inherits a Hermitian metric from the classical Hilbert-Schmidt inner product. The renormalized trace tr ∆ extends the classical trace tr of trace class operators to a smooth linear functional on Cl(S 1 , V ). We investigate here the possible (maybe naive) extension of the classical Hilbert-Schmidt inner product to Cl(S 1 , V ) via tr ∆ .

3.1.
Calculation of renormalized traces. Let (z k ) k∈Z is the Fourier L 2 −orthonormal basis. Let us recall that there exists an ambiguity on ǫ(D) concerning its action on z 0 , which can be, or not, in the kernel of p + , or in the eigenspace of the eigenvalue 1 or −1. Depending on each of these three possibilities respectively, we set ǫ(k) as the eigenvalue of ǫ(D) at the eigenvector z k .  Proof. We have here even-even class operators, so that the renormalized trace is commuting in all items.
(2) Here the functions u and v are real-valued, which means that one should consider the real Fourier basis for the summation. However, since the real Fourier basis is a linear combination of the complex one, we investigate first the renormalized trace with u = z n and v = z m . Then XY * (z k ) = (−(m − k) 2 )z n−m+k . Then we adapt the previous computations: . By the way, passing from the complex Fourier Basis to the real Fourier basis, (3) Since a and X are even-even class, we thave that tr ∆ (aX) = tr ∆ (Xa). Setting a = z n and X = u d dx , with u = z m , we compute By the way, tr ∆ (aX) = tr ∆ (Xa) = 0.

3.2.
Extension of the Hilbart-Schmidt metric to F Cl.. Proof. We proceed set by set, following the order of the statement of the Theorem. On Cl(S 1 , V ): The formula tr (AB * ) gives obviously a sesquilinear form. We prove first that the form is Hermitian: Let (A, B) ∈ Cl(S 1 , V ). tr ∆ (BA * ) = tr ∆ ((AB * ) * ) = tr ∆ (AB * ).
Let us then prove that it is non-degenerate. Let A ∈ Cl(S 1 , V ), let u ∈ C ∞ (S 1 , V )∩ (ImA − {0}) which is the image of a function x such that ||x|| L 2 = 1, and let p x be the L 2 − orthogonal projection on the C−vector space spanned by x. Then, let (e k ) k∈N be an orthonormal base with e 0 = x. For Re(s) ≥ 2ord(A) + 2, we observe, applying commutation relations of the usual trace of trace-class operators, the following: By the way, the meromorphic continuation to C of s → tr (Ap x ) * ∆ −s A exist and coincide with the meromorphic continuation of s → tr A (Ap x ) * ∆ −s , in particular at s = 0.
By the way, since lim s→0 ∆ −s/2 = Id for weak convergence, The operator Ap x is a smoothing (rank 1) operator, and hence is in Cl(S 1 , V ), which ends the proof. On Cl ee (S 1 , V ) and on Cl(S 1 , V ) ⊕ V ect(S 1 ): The same arguments as before hold for non-degeneracy, both on Cl ee (S 1 , V ) and on Cl(S 1 , V ) ⊕ V ect(S 1 ). The rest of the arguments follow from the inclusions Cl ee (S 1 , V ) ⊂ Cl(S 1 , V ) and Remark 3.3. We remark that (., .) ∆ is bilinear, non degenerate but not positive. Indeed, from relation (1) of Lemma 3.1, C ∞ (S 1 , M n (C)) is an isotropic Lie subalgebra for (., .) ∆ which proves that this R−bilinear symmetric form is not positive.
From the Lie algebra Cl(S 1 , V ) ⊕ V ect(S 1 ), we then span by right-invariant action of F Cl * (S 1 , V ) on T F Cl * (S 1 , V ) a right-invariant pseudo-metric. For this goal, the Lie algebra elements are identified as infinitesimal paths, and actions and Lie brackets are those derived from the coadjoint action (and right-Lie bracket) of F Cl * (S 1 , V ) on Cl(S 1 , V ) ⊕ V ect(S 1 ), while we consider the trivial mapping defined by the sum Cl(S 1 , V ) ⊕ V ect(S 1 ) → Cl(S 1 , V ) = Cl(S 1 , V ) + V ect(S 1 ) in order to compute Re(.; .) ∆ . The same constructions hold for the pseudo-Hermitian metric (.; .) ∆ on Cl * (S 1 , V ). Definition 3.4. Let A ∈ F Cl 0, * (S 1 , V ) and let a ∈ Cl 0 (S 1 , V ) ⊕ V ect(S 1 ). We note by R A (a) the (right-)action by composition Then, identifying T A F Cl 0, * (S 1 , V ) with R A Cl 0 (S 1 , V ) ⊕ V ect(S 1 ) we set a smooth pseudo-Riemannain metric on T F Cl 0, * (S 1 , V ) by defining for and hence for (R A (a), R A (b)) ∈ T A F Cl 0, * (S 1 , V ) 2 ,

4.
In search of pseudo-Hermitian connections for (., .) ∆ There exists some difficulties in describing the whole space of connection 1-forms Ω 1 (F Cl Dif f (S 1 ) (S 1 , V ), Cl(S 1 , V ) ⋊ V ect(S 1 )). Indeed the space of smooth linear maps acting on Cl(S 1 , V ) is actually not well-understood to our knowledge. In particular, finding an adjoint of the adjoint map for (., .) ∆ fails apparently due to the non-traciality of tr ∆ . We consider here a class of connections where this smooth linear endomorphism is defined by composition by a smoothing operator. The resulting technical simplifications enables us to get pseudo-Hermitian connections for (., .) ∆ . Most of them can be easily adapted to get pseudo-Riemannian connections for Re(., .) ∆ .

4.1.
A class of connections. Let us define now, for w ∈ Cl(S 1 , V ) such that .
Let us analyze the connection Θ w with w = iǫ(D).

4.2.
Pseudo-Hermitian connections associated with a skew-adjoint pseudodifferential operator. Let w ∈ Cl(S 1 , V ) such that w * = −w. For example, one can consider the example w = iǫ(D).
Let us now analyze Theorem 4.4. θ w is the connection 1-form of a pseudo-Hermitian connection of (., .) ∆ .
Proof. Let (a, b, c) ∈ Cl(S 1 , C) 3 . θ s,r a = Θ s,r a b − Θ s,r a * b = bs(a − a * )s * . Lemma 4.6. ∀a ∈ Cl(S 1 , V ), ∀s ∈ Cl −∞ (S 1 , V ), • Θ s,l a * is the adjoint of Θ s,l a for (., .) ∆ • Θ s,r a * is the adjoint of Θ s,l a for (., which proved the first point. By the way, given A ∈ F Cl Dif f (S 1 ) (S 1 , V ), if the phase diffeomorphism g of A is orientation preserving, then, under the blockwise decomposition H + ⊕ H − , A +− and A −+ are smoothing operators according to [35], and if g ∈ Dif f − (S 1 ), A ++ and A −− are smoothing operators. where Ω iǫ is the curvature of Θ iǫ .