International Journal of Geometric Methods in Modern Physics
c World Scientific Publishing Company
arXiv:1705.02974v1 [math-ph] 8 May 2017

Differential calculus on manifolds with a boundary. Applications.

1 Dipartimento di Fisica “E. Pancini” dell Universitá “Federico II” di Napoli, Complesso
Universitario di Monte S. Angelo, via Cintia, 80126 Naples, Italy.

2 Sezione INFN di Napoli, Complesso Universitario di Monte S. Angelo, via Cintia, 80126
Naples, Italy.

3 Turku Center for Quantum Physics, Department of Physics and Astronomy, University of
Turku, FIN-20014, Turun yliopisto, Finland.

This paper contains a set of lecture notes on manifolds with boundary and corners, with
particular attention to the space of quantum states. A geometrically inspired way of
dealing with these kind of manifolds is presented, and explicit examples are given in
order to clearly illustrate the main ideas.

Keywords: quantum mechanics, open quantum systems, manifolds with boundary


1 Introduction 2

2 Natural Constructions and Differential Calculus on Vector Spaces 11
2.1 Tensors on a vector space and on its dual . . . . . . . . . . . . . . . 12
2.2 Vector spaces with additional structures . . . . . . . . . . . . . . . . 13
2.3 Homology Operators on Multivectors . . . . . . . . . . . . . . . . . . 14

3 Manifolds 17
3.1 Brackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Clifford Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Clifford Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4 Manifolds with boundaries and corners 28

4.1 The geometry of OA and DA . . . . . . . . . . . . . . . . . . . . . . 30
4.2 Kossakowski-Lindblad vector fields . . . . . . . . . . . . . . . . . . . 37

5 Conclusions 39

References 40



2 Contents

1. Introduction
This paper contains a revised and extended version of the lectures given by one of
us (G.M.) at the “International Workshop on Quantum Physics: Foundations and
Applications” held at the Centre for High Energy Physics of the Indian Institute for
Science of Bangalore. It contains a leisurely introduction to the differential calculus
on manifolds with boundaries. Its main motivation is differential calculus on the
space of quantum states. It deals with a geometrically inspired way of viewing and
solving problems about subalgebras of smooth functions and quotient manifolds.
The main idea is to investigate a manifold M by means of the algebra of function
F(M ). Actually, we shall argue that when it comes to the quotient manifolds and
submanifolds it is convenient to consider also the full algebra of tensor fields on
M , say T (M ). The differential aspects of the manifold are mainly carried by the
Lie algebra of derivations (vector fields) of F(M ). The style is rather informal. We
emphasize examples, motivations and intuition and refer to the literature for full
proofs and the most general definitions.
Manifolds with boundaries arise quite naturally in the description of realistic
physical systems, both in classical and quantum physics. Typical examples are:

• Biliards, classical and quantum;
• Electrons moving in wires, Hall effect;
• Wave guides;
• Source singularities.

The most simple example of a manifold with a boundary is the semiline R+ , i.e.,
the real numbers a ≥ 0. In general, the Cartesian product of manifolds with bound-
ary is not a manifold with boundary, but a manifold with corners. For instance, the
Cartesian product of N semilines is the prototype of manifold with corners.
In general, manifolds with boundaries arise every time we divide the carrier space
into an accessible region and its complement which we assume to be not accessible
to observation. Thus, what happens on the boundary is an effective way to represent
whatever happens “on the other side” which has an effect on the accessible part.
Most of the manifolds with a boundary we are going to be interested in arise
as reduction of homogeneous spaces under specific equivalence relations. Thus, in
principle we shall assume that manifolds with boundaries may be ”unfolded” to
manifolds without boundary. As it is usually the case, the unfolding procedure
is not uniquely defined, therefore, other considerations will enter the ”unfolding”
procedure, like symmetry considerations, minimality, simplicity, and so on.
In the context of C ∗ -algebras, this manifold may represent the space of states
of a finite-dimensional, commutative C ∗ -algebra A. Consequently, it represents the
space of physical states of a finite-dimensional classical system.
In general, boundaries are generated by inequalities, the most common inequal-
ities emerging in physical situations are associated with non negativity (energy,
entropy production, probabilities and so forth).


Contents 3

To be a little more formal, we start by considering the simplest carrier space,
i.e., a vector space V . Let α ∈ V ∗ be an element in the dual space of V , and
consider the map α : V → R. The inverse image of R+ is a half-plane, say V + , and
the inverse image of 0 ∈ R will be a hyperplane. Using the same procedure with
a set of linearly independent forms on V we obtain what is known as a quadrant,
or simplex. The quadrant is the model space for the category of manifolds with
corners ([1]).
Note that the procedure outlined above for finite-dimensional vector spaces
makes sense also in the more general context of Banach Spaces.
Manifolds with corners very often arise as quotients by action of discrete group,
that is, as orbifolds. For instance, the half-plane may be considered also to be
the orbifold of the vector space V under the action of Z2 , (x ; y) 7→ (−x ; y). The
quotient space would be diffeomorphic with V + . In intrinsic terms we would define
an equivalence relation in V by setting v1 ∼ v2 if |α(v1 )| = |α(v2 )|, i.e., either
α(v1 ) = α(v2 ) or α(v1 ) = −α(v2 ).
This is interesting because, often, it is possible to “double” a manifold with
a smooth boundary in order to unfold the quotienting procedure. Consequently,
we can discuss multi-differential operators and higher order differential operators
directly on a manifold without a boundary and then apply a specific reduction
procedure. This should be kept in mind when discussing self-adjoint extensions and
Dirac operator on manifolds with boundary.
Another interesting example comes from Quantum Mechanics. The tomographic
approach allows us to associate to every state ρ a probability distribution pρ with
respect to a specific resolution of the identity. For a finite level system, the space
of probability distribution becomes an n-simplex, where n is the dimension of the
Hilbert space of the system (an example is shown in Fig.1).

Fig. 1. Simplex representing the space of probability distributions of a three-level quantum system

This is clearly a manifold with corners. The reader may consider the book [2]
for additional aspects connected with Quantum Mechanics.
More generally, manifolds with boundary or corners are often associated with
reduction procedures of dynamical systems, or actions of Lie groups or Lie alge-


4 Contents

bras. These procedures naturally lead to consider quotient spaces associated with
dynamical systems. It could be the space of orbits or the space of leaves when more
than one vector field is involved, for instance a family of vector fields closing on a
Lie algebra. Unfortunately, quotient spaces can be wild objects in the sense that
they could lack of a differential structure as a smooth manifold, or, even worse,
they could present pathological topologies, e.g., non-Hausdorff topologies. In all
these cases it would be difficult, if not impossible, to recover the notions of vector
fields, differential forms, and tensor fields defined on the quotient space.
From a mathematical point of view, a convenient approach to deal with quotient
spaces is the algebraic one, in which a manifold M is replaced by the algebra
F(M) of functions on it. Indeed, the algebra F(N ) of functions on a quotient space
N ∼ = M/ ∼ can be identified with a subalgebra of the algebra of functions F(M)
given by functions that are constant on the equivalence classes. More generally,
when N admits a suitable differential structure, the algebra of covariant tensors on
N can be represented as an invariant subalgebra (under the equivalence relation) of
the space of covariant tensors on M. However this point of view can be used also in
the cases in which N presents no differential structures. Indeed, in such a situation
we are still able to define an invariant subalgebra of the algebra of tensor fields on
M which we may declare to be the algebra of tensor fields on N . Furthermore, this
framework allows to replace the notion of vector field with the more general concept
of derivation of an algebra, since vector fields on a smooth manifold are derivations
of the algebra of smooth functions with respect to the pointwise product ([3]).
In order to clarify these ideas, we will give some concrete examples. Let us start
with the semiline R+ 0 ; it can arise by reduction from the real line R with respect
to the action x 7→ −x of the discrete group Z2 . This space is the prototype of
manifold with corners, and thus its differential structure is slightly more complex
than that of a differential manifold. From the algebraic point of view, this space can
be described using the subalgebra F+ ⊂ F(R) of “even” functions on the real line.
Not all the derivations of F(R) are derivations of F+ . Indeed, since the derivative
with respect to x of an even function is an odd function, the module of derivations

of F+ is generated by x ∂x .
For differential manifolds, we are used to the idea that, at least locally, all
covariant tensors can be constructed out of functions and their differentials, but
on the semiline, for instance, the tensor dx ⊗ dx, globally well defined, cannot be
generated by differentials of even functions.
A less trivial example comes from the two dimensional torus. Specifically, let us
consider T2 ∼ = S 1 × S 1 and the vector field X = a Xθ + b Xφ , where Xφ , Xθ are
the canonical vector fields on the circles composing the torus, and a/b an irrational
number. It is well known that the quotient space with respect to this dynamical
flow can be identified with the discrete set of irrational number, and the quotient
topology on this set is then the discrete topology. It is clear that there is no dif-
ferential structure other than the trivial one. However, writing αθ , αφ for the one
forms that are dual to Xθ , Xφ respectively, we immediately find that b αθ − a αφ

The quotient space that arises from the reduction with respect to this dynamical flow is a non-Hausdorff space. red curves are the corresponding dynamical orbits on S 1 . 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 5 is a basis for a one-dimensional module of differential one forms that are invariant with respect to the dynamical flow. The peculiar property is that the ”quotient” would have forms and covariant tensors. May 9. Since the energy function on the phase space is a constant of the motion. the situation can be handled if we consider functions and forms which are invariant under the equivalence relation. this space presents a topology with which it is difficult to work with. represented in the z-direction. A different pathology arises for instance with the space of orbits of the simple pendulum. whose elements are all those functions in F(M) that depends on the energy. 2. According to what has been said above. there are two different kinds of orbits: the orbits which are circulated clockwise. The phase space of this dynamical system is the cotangent bundle T ∗ S 1 over the manifold S 1 . . Apart from the absence of a differential structure. If one looks at the phase diagram. It can be seen that. only vector fields which end up being tangent to the boundary are con- sidered. Fig. when E = E? there are three orbits. it is possible to identify three different families of orbits: up to a certain value E? of the energy E. for E > E? . we could think of this module as the space of differential one forms on the quotient space. The diffeomorphism group splits into the diffeomorphism group of the interior and the diffeomorphism group of the boundary. but they would not be generated by differentials of functions defined on the quotient space. what may be called the transversal part is usually assumed to vanish on the boundary. Folding of the phase space T ∗ S 1 and its quotient with respect to the dynamics. For each value of the energy. we can describe the quotient space by means of the invariant algebra FH . in the mathematical literature. and the other two are bounded but not closed. However. when E > E? the orbits can be divided into two classes according to the orientation of the actual motion. one is a single point. parametrized by energy and phase. the orbits are closed circles in the phase space. when dealing with a manifold with a boundary. Often. from the algebraic point of view. and the orbits which are circulated counterclockwise.

3). Fig. Fig. in red are drawn gradient motions with respect to n̂-axis. This space can be seen as a two-dimensional sphere embedded into a three-dimensional Euclidean space. The quotient space with respect to this dynamical flow can be identified with a closed interval of the real line (see Fig. let us introduce a second axis n̂ such that n̂ · ẑ 6= ±1. The gradient motions are generated by vector fields Xf such that iX g = df with f a smooth function on the sphere. Let us consider a rotation along the ẑ-axis. on the sphere. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 6 Contents From the physical point of view. This kind of dynamical evolutions often results in the motion from the boundary to the interior. 4. . by considering a specific orbit on the sphere. Now. where the boundary points are the equivalence classes associated with the north and south poles respectively (fixed points). however. Bloch sphere relative to a two-level quantum system. we might be interested in the descrip- tion of dynamical evolutions on manifolds with boundaries or corners. its projection would give a The Euclidean metric δ on the three-dimensional determines a rotationally-invariant metric g on the sphere. In particular. Representations of the space of rotations around the ẑ-axis on the Bloch sphere. the so-called gradient motionsa with respect to the n̂-axis are precisely the meridians on the sphere with respect to the n̂-axis (see Fig. In red is drawn a gradient motion which moves the boundary towards the interior. In green are represented rotations around the ẑ-axis. It is well known that. For example. let us consider the space of pure states of two-level quantum sys- tem (qubit). May 9. 3.4).

k}. ∂y . If we want to describe dynamical systems on these manifolds.. . R. (2) A simple example is a Ball: x21 + x22 + · · · + x2k ≤ 1 or 1 − (x21 + x22 + · · · + x2k ) ≥ 0. y) on R × R+ + 0 . ∂y∂ j j ∈ {1. they may be thought of as arising from constraints: f1 (m) = 0. (4) Manifolds with boundaries can be always thought of as an “open part” M̊ ⊂ M and a “boundary part” ∂M ⊂ M . y) ∈ R2 | 0 ≤ x ≤ a. under mild regularity requirements. May 9.. On a manifold with corners we would have xl ∂xl l ∈ {1. with N sufficiently large. Singer (APS) index theorem on manifolds with boundary. fk (m) = 0. . on the quotient space. . manifolds may be always described as submanifolds of some RN . x3 this would be the Bloch ball for the quantum states of a q-bit. When there are corners. the boundary points are mapped in the interior. (3) For x1 . If for simplicity we introduce Cartesian orthogonal coordinates (x . For instance. instead. . ∂M is not a smooth submanifold. When necessary. . n − k}. . (1) Manifolds with a boundary arise. it is possible to move from the poles of the ẑ-axis. we can consider the algebra of vector fields x ∂x . a ∈ R. fk (m) ≥ 0.. These vector fields are complete and generate an action of an Abelian Lie group. . let us con- sider the 2-dimensional example R × R+ 0 . f2 (m) = 0. 0 ≤ y ≤ a}. f2 (m) ≥ 0. we need to consider their tangent or cotangent bundles: T M for Newtonian and Lagrangian mechanics T ∗ M for Hamiltonian mechanics. from inequalites: f1 (m) ≥ 0 .. ∂ . In his proof of the Atiyah. . By means of the Whitney’s theorem. M itself may be thought of as immersed in some open larger manifold of the same dimension. y for R0 nand x foro the transversal part ∂ ∂ to it.. .. and thus. Melrose has developed a differential calculus and a b-geometry starting with an algebra of complete vector fields ([4]). Let us make some general considerations where we formalize some of the aspects we have been mentioning. . When multiplied by functions on R+ 0 they define a Lie module. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 7 rise to a motion with a fold. x2 . 2. In this respect. A billiard would be given by a submanifold in R2 defined by: B = {(x. Patodi.

first order differential operators appear as a “gauged” version of vector fields (derivations). the fibers at the boundaries must have the same dimension they have at the interior. this structure is not enough. we can construct the holomorph Lie algebra ([5] page 523). as we have already stressed. However. It is clear that. this is unsatisfactory from the physical point of view. In addition to all those vector fields that vanish on the boundary we need at least one that does not. 5). Essentially. although this occurs often in the mathe- matical literature. one is the boundary and the other is the full interior. if we want to consider the case in which n → ∞. Say I ⊂ R the interval: I = {x ∈ R. b For instance. the potential becomes a ”square well” (see Fig. in the case of a manifold with boundaries. Indeed. the standard picture of the tangent bundle requires qualification. Additionally. |x| ≤ 1} (6) which is considered as the configuration space for the dynamical system described by this Lagrangian function: 1 1 L= mv 2 − 1 . Indeed. . It means that the velocity at the boundary can be only zero and it is well accomodated by the phase portrait. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 8 Contents With the vector fields x ∂x ∂ ∂ and ∂y . To accomodate this situation. ∂ ∂ The orbits generated by x ∂x and ∂y can never connect a point on the boundary with a point in the interior. and ∂y + ∂yj we get first order differential operators out of derivations. or T ∗ M̊ and T ∗ (∂M ). To make clear this point let us introduce an example. (5) and the sections of T (∂M ). we can construct the associative algebra of all differential operators. n ∈ N. otherwise we can not move from the boundary towards the interiorb . in this case we will have: T M̊ and T (∂M ). May 9. because dechoerence processess do not preserve the rank of quantum states. are not enough to implement the motion from the boundary to the interior. this would mean that we are not able to describe dechoerence of a pure state. and we can either construct multidifferential operators. the diffeomorphisms generated by the algebra of vector fields preserve the boundary. the tangent bundles of these two manifolds ˚ have dimensions dim T I = 2 and dim T (∂I) = 0. i. considered as derivations of F(R+ 0 ). 6) and on the boundary all the values of the velocity are allowed (see Fig. However. There are two orbits of the Abelian group. the vector fields tangent to the boundary.e. in this limit. (7) 2 (1 − x2 ) n If we now split the interval as ˚I and ∂I. in the quantum case. by noticing that with the similarity transformations e−f x ∂x ∂ ∂ ef = x ∂x + x ∂f ∂ ∂f ∂x . or multivector fields.. that is.

→ Q̃. For a given value of the total energy. (9) b∈∂Q which allows the boundary to have the same fiber of the inner pointsc . the vectors of this fiber bundle are not all tangent to the boundary. (8) In other words. let us consider. the space of positions and allowed velocities will not be a manifold with boundary anymore. which generates a semigroup of transformations on the half-plane in the sense that. The paper is organized as follows. May 9. by means of the immersion ∂Q . 5. we can induce the pull-back bundle: i∗ (E(Q)) = {(q 0 .5 1. in the limit n → ∞. we will review the basic tools of differential calculus starting with the case of a vector space V . considering the generated flow {φτ } on the ambient space. First of all. there are some of them that allow to move the points from the boundary towards the interior. With this restriction. let us consider the vector π i bundle E −→ Q̃. therefore. we have to restrict only to motions which take place in the half-plane. and then passing c Obviously. the same procedure can be carried on for the contangent bundle T ∗ Q and momentum p. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 9 vHxL 4 2 x -1. ∂ Returning to the example of the half-plane. Moreover. in a more general fashion. Then.0 -2 -4 Fig. an open manifold Q̃ which includes a manifold with boundary Q. in general. namely Q ⊂ Q̃. A possible way to achieve this situation is by means of a pull-back bundle.0 -0.5 0. Accordingly. . e) ∈ ∂Q × E | i(q 0 ) = π(e)} ⊂ ∂Q × E. Indeed. we would find that the velocity at the boundary would jump from v to −v. it is the union of the fibers: [ Eb (Q). its general action would take us out of the half-plane. we consider the vector field ∂x .

Notice that the axes are centered in (0. On this space it is possible to define a Clifford algebra structure by means of the so-called ∨-product.2 x -2 -1 0 1 2 Fig. 1) to the more general case of a manifold without boundary. The algebraic formalism allows to immediately generalize the results of this section to the more general case of smooth manifolds by replacing scalars with scalar smooth functions.0 1. As shown above. This is an example of a stratified manifold with boundary. The space DA will always be ∗ thought of as a subset of a linear space. we will show how to implement the notion of a vector field ΓL which is transversal to the strata of DA . so that we are be able to describe some interesting physical phenomena that do not preserve the rank of quantum states. being made up of different strata each of which is a differential manifold on its own. 6. we will outline the main features of differential calculus in an algebraic setting. essentially. . dechoerence or ∗ damping phenomena. this allows us to bypass some of the technical problems due to the absence of a global differential structure on DA . (7) will describe a free particle of mass m which collides on the boundary and inverts its motion. The differential structure of different strata will be. as a ∗ derivation of F(OA ). but. section 2 will be devoted to natural constructions on vector spaces. in section 4 we will consider the space of states DA of a finite-level quan- tum system associated with a finite-level C ∗ -algebra A. on the so-called expectation value functions. Consequently. this is extremely useful in dealing with unfolding and reduction procedures. In particular. According to the ide- ology exposed in this introduction. The final step is enlarging the algebra of smooth functions to the exterior algebra of differential forms.. By using the duality between a manifold and the algebra of smooth functions on it. so that a global differential structure in some ambient space will always be at our disposal. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 10 Contents V HxL 2. e. In the limit n → ∞ the dynamical system defined by Eq. the differential structure inherited from this ambient space.4 1. Next. that is. a topological space with boundary hav- ing no differential structure as a whole. We will review the differ- ential geometry of this space as it has been done in [6]. May 9. rather.8 1.6 1. namely. the dual space OA of the space of observables. This vector field ΓL is defined on OA and the fact that it is transversal to the strata of DA is implemented by looking at its action.g. This can be very useful in relation with the physics of spinors and Dirac operators.

say F(V ). it means that for each m ∈ M. . (13) Similarly we can build multilinear maps. not restricting only to linear transformations. being functions they can be multiplied point-wise and provide us with an associative. Natural Constructions and Differential Calculus on Vector Spaces Before embarking in the differential calculus on manifolds with a boundary it is con- venient to review differential calculus on manifolds without boundary. K) with K the field of real or complex numbers. may also be exported on a generic manifold when they are glued together as [ Vm . say Um . this is the space of linear functions on V . May 9. we should allow for general nonlinear transformations. These linear functions may also be multiplied differently. It is useful to step back and recall that M may be thought of as a generalization of a vector space V . v) := f (u)g(v) . there is an open neighbourhood of m. we may consider M to be a vector space V or an affine space having V as a model. commutative algebra with identity. which contains all polynomials built out of V ∗ . diffeomorphic to V or. All geometrical or algebraic structures we may build out of V . with the following properties: i) There exists only one point v0 ∈ V such that E(v0 ) = 0. for simplicity. M can be recovered as the space of algebra homomorphisms HomK (F. said differently.8] and [9]. the pointwise product is: (f · g)(v) := f (v)g(v) . in one-to-one correspondence in a smooth way to V (or an open contractible neighbourhood of the origin). For further details we refer to [7. The advent of non-commutative geometry in physics has stressed the role of the identification of a manifold with the algebra of functions defined on it: M F(M). 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 11 2. (11) m∈M We first notice that along with V we have V ∗ ≡ LinK (V. Thus. (10) Where F is given. without any further requirements. A tensorial description of the vector space structure of V is provided by the homogeneous first order differential operator E. K). If we have in mind the transition to man- ifolds. (12) but also as providing a function to be evaluated on pair of vectors: (f × g)(u . called Euler operator.

therefore the required completeness implies that the open neighborhood should actually give rise to the full vector space.. V coincides also with the linear function defined on V ∗ : V = LinK (V ∗ . its flow defines a one-parameter group of trans- formations. e2 . for instance E = (xj +cj ) ∂x j . considered as an algebra of functions. en ) such that ej (ek ) = δkj . and we set xj (v) = ej (v). . Tensors on a vector space and on its dual In finite dimensions. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 12 Contents ii) E is a complete vector field. . K). iii) E f = 1 · f has n independent solutions if n = dim V . e2 .1. (15) Since this characterization is given in terms of a vector field. . From this point of view F(V ) has some ”universal properties” because it only depends on the differential properties of V and does not depend on its linear structure (see [5]). It is clear that F(V ). we define the dual basis (e1 . May 9.e. By means of the third requirement. it will have all the properties of a ”linear structure” and therefore a new polynomial subalgebra of F(V ). . in any linear coordinate systemd for V the first order differential operator is: ∂ E = xj . iv) E f = 0 · f has only constant solutions. v ∈ V . this description does not depend on the choice of a particular coordinate system. is defined by: (f1 + f2 )(v) = f1 (v) + f2 (v) (f1 · f2 )(v) = f1 (v)f2 (v) (14) h (f1 + f2 ) = h · f1 + h · f2 and therefore contains no reference to the vector space structure of V . . . Notice that here linear independence coincides with functional independence. 2. Notice that conditions ii). en ) a basis of solutions of iii). . Indeed. i. This infor- mation will be carried by the Euler operator E which allows to define homogeneous polynomial functions of degree k by means of the following eigenvalue equation: E(f ) = k f. . we can select n independent linear func- tions and use them to define the vector space structure. . (16) ∂xj ∂ If we now select a different Euler operator. V = (V ∗ )∗ . (17) d Say (e1 . iii) and iv) would be satisfied on some open neighbor- hood of v0 .

say: MLin (V × V ) ≡ V ∗ ⊗ V ∗ (23) MLin (V × V × V ) ≡ V ∗ ⊗ V ∗ ⊗ V ∗ = (V ∗ ⊗ V ∗ ) ⊗ V ∗ and so on. We denote by: T 0 (V ) ≡ K. T 0 (V ) = V ∗ . Occasionally. and thus. Vector spaces with additional structures Assume V carries a skew-symmetric binary. ] : V × V −→ V (24) such that: e Fromnow on we will assume that all the vector spaces are defined on the same common field K. We should pay attention to the fact that multinear maps are not linear maps. The tensor product ⊗ defines an associative algebra structure on tensors. we shall denote them by S(V ∗ ) and Λ(V ∗ ). for the sake of notational simplicity. . T σ (V ) ∼ = Tσ (V ∗ ) (21) or.2. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 13 In infinite dimensions this relation is not guaranteed and when it holds we say that V is totally reflexive. These multilinear maps are also called tensors and we denote them by: T σ (V ). From linear mapse Lin(V ) we can generalize to multilinear maps: MLin (V × V × · · · × V ) (18) and dually: MLin (V ∗ × V ∗ × · · · × V ∗ ) (19) along with multilinear mixed maps: MLin (V × · · · × V × V ∗ × · · · × V ∗ ). 2. It contains two subalgebras of totally symmetric tensors and totally skew-symmetric tensors. May 9. bilinear product denoted by a square bracket: [ . we will drop the indication of K. (22) We shall also write as a product. (20) Remark 1. Tρσ (V ). in the mixed case.

û}ŵ + û{v̂. v3 ] . −→ Λ1 (V ) −→ 0 (28) defined by: X ∂(v1 ∧ v2 ∧ · · · ∧ vp ) = (−1)i+j+1 [vi . We have. it is derived from the Jacobi identity. Before en- tering these aspects. v2 ] = −[v  2 . Homology Operators on Multivectors We have ([10]) a sequence: ∂ ∂ ∂ Λp (V ) −→ Λp−1 (V ) −→ . ûŵ} = {v̂. commutative product on the point-wise algebra F(V ). dû)(α) = α [v. [ . u] (25) then {v̂. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 14 Contents i) [v1 . Property ii) says that this operation defines derivations for the binary. let us remark that any Lie algebra structure on V induces a Poisson bracket on F(V ∗ ). u] (27) ∗ where α ∈ V . . It is usually called the Jacobi identity.3. It is possible to develop an exterior differential calculus at a purely algebraic level. (29) i<j .v1 ]    ii) v1 . v3 + v2 . This is done first on linear functions by exploiting the identification V ≡ Lin (V ∗ ) and then extended by using the derivation property with respect to the associative. the algebraic differential calculus becomes richer and contains all essential ingredients of the usual differential calculus on manifolds. . if we denote by v̂ the linear function corresponding to v [ {v̂. ŵ}. 2. [v1 . ]) acts as an algebra of derivations on F(N). In the few coming sections we shall illustrate these various aspects. v3 ] = [v1 . This pairing defines a Lie algebra structure. When in addition the Lie algebra (V. and is the counterpart of d · d = 0 on forms ([8]). This bivector field satisfies a homology property when applied to higher multi- vectors. û} = [v. v2 ]. May 9. [v2 . we can define also a controvariant tensor field on V ∗ by setting:  Λ(dv̂. bilinear product. (26) As v̂ and û are (linear) functions. with N any smooth manifold. vj ] ∧ v1 ∧ · · · ∧ vˆi ∧ · · · ∧ vˆj ∧ · · · ∧ vp .

Given any endomorphism ([13]) T : V −→ V (37) we consider the natural extension to Λ(V ) by setting X δT : v1 ∧ v2 ∧ · · · ∧ vp 7−→ v1 ∧ v2 ∧ · · · ∧ T (vj ) ∧ · · · ∧ vp .h. (36) This bracket is the Schouten-Nijenhuis bracket ([11. We remark that ∂ is not a derivation with respect to the wedge product. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 15 where the ˆ notation means that the corresponding vector is missing from the ex- pression. It follows that ∂ · ∂ = 0. We should remark that the r. (38) j . u]. V ) (34) by: v 7−→ adv (35) with adv (u) = [v. H] = ∂(G ∧ H) − ∂G ∧ H − (−1)q G ∧ ∂H. (32) It is now possible to define: Lu = u ∂ + ∂u (33) which turns out to be the extension to Λ(V ) of the adjoint representation ad : V −→ Lin (V.s. H ∈ Λp (V ) we define the extension of the Lie algebra product on V to the algebra of multivectors by setting: (−1)q+1 [G. May 9. We have also v : Λp −→ Λp+1 (30) by: v : v1 ∧ v2 ∧ · · · ∧ vp 7−→ v ∧ v1 ∧ v2 ∧ · · · ∧ vp (31) with the property v u = −u v . Given two multivectors G ∈ Λq (V ).12]) which makes Λ(V ) a graded Lie algebra. appears as the “deviation” from a graded derivation of ∂.

(39) Now we combine these operations to get: ∂T = δT · ∂ − ∂ · δT : Λ(V ) −→ Λ(V ). 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 16 Contents For T = 1. α)|(u. iv Ω0 = 0 (44) with the following properties: iv iu + iu iv = 0. v2 . v3 ) ∈ V (41) defines a tensor associated with T which we might call a Nijenhuis tensor. Now we want to dualize this construction. v2 ] .  dα(v1 ∧ v2 ) = α [v1 . we simply get: δ1 : v1 ∧ v2 ∧ · · · ∧ vp 7−→ p(v1 ∧ v2 ∧ · · · ∧ vp ).v]. (47) We may now define pairings on V ⊗ V ∗ by setting: (v. This tensor vanishes iff ∂T · ∂T = 0 and is a boundary homology operator. (43)  dβ(v1 ∧v2 ∧· · ·∧vp+1 ) = i<j Again we define a contraction iv : Ωp −→ Ωp−1 . say α ∈ V ∗ . iv d + div = Lv . May 9. (40) In general ∂T · ∂T 6= 0 and we have: ∂T (∂T (v1 ∧ v2 ∧ v3 )) = NT (v1 .u] . (45) Lv will be called the Lie derivative and we get: [iv d + div . β) + = α(u) + β(v) (48) . (42) This time d · d = 0 because of the Jacobi identity and on a generic antisymmetric multilinear map we get X (−1)i+j β [vi . vj ]∧v1 ∧· · ·∧vi ∧· · ·∧vj ∧· · · ∧p+1 . Lv ] = L[u. We first set for any linear function on V . iu ] = i[v. (46) Moreover [Lu .

2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 17 or (v. f2 )] = ([D1 . that is the subgroup of automorphism which not only respect the vector space structure of F. (49) It is possible to define a pairing between Λ(V ∗ ) and Λ(V ) by setting: v1 ∧ v2 ∧ · · · ∧ vp |α1 ∧ α2 ∧ · · · ∧ αp = det ||αj (vk )||. In this section we have shown how to construct differential calculus at purely algebraic level. in the sense that there was no need to introduce the concepts of points. and quotients to define derivatives. the field of coefficients. A dualization. all of our constructions are glued to- gether by taking the union over points of M. T ∗ M. 3. Essentially. α)|(u. and the jet bundle J k (M). re- placing M with F(M). Manifolds It is possible to move all our previous constructions from a vector space V to a manifold M. This semidirect product is a very large infinite dimensional group which contains the semidirect extension Diff(F) n F as a subgroup. sections of T M will be replaced by Der(F(M)). D2 ] . Now we can construct T M. In particular this would give a Laplacian or a D’Alambertian operator when we move from vector spaces to modules and the Lie algebra is realized in terms of vector fields. The group Diff(M) for any differential manifold says that all points look alike. either real or complex valued. f1 ). the derivations of F(M). is possible. With this construction all tensors become tensor fields. Tsr (M). In the next section we will make this statement more precise. When V itself carries an inner product. (50) With respect to this pairing d and −∂ are the transpose one of the other. It is possible to consider F(M) as a trivial Abelian Lie algebra and construct the Holomorph of F by considering the semidirect product DerF n F. If F is thought of as an Abelian group we can consider Aut(F) × F. D1 f2 − D2 f1 ). therefore. limits. iv is the transpose of v . each contractible open neighbourhood of a point will be diffeomorphic to its tangent space. but also the pointwise associative and commutative product on F. [(D1 . May 9. It is also possible to construct an Hodge-∗ operator which allows to build second order “differential operators” out of the introduced exterior derivative. real or complex numbers will be re- placed by the ring F(M) of functions on M. As an example of an automorphism of F as an Abelian group which is not an automorphism of F as an algebra we can consider e−it∆ : L2 (M) → L2 (M) . β) − = α(u) − β(v). it is possible to induce an inner product on the associated tensor spaces. (D2 .

f2 . In order to make clearer this point let us consider a simple example. i. Derivations of the algebra F(M) of functions on the manifold M. Even if Gelfand- Naimark theorem states that every commutative C ∗ -Algebra can be represented isometrically as the algebra C 0 (M) of continuous functions on a suitable manifold M. represents the algebra of differential operators acting on F. Let R the real line and F(R) the algebra of functions on R equipped with the Abelian product: Z +∞ (f ? g)(p) = f (p − q)g(q)dq −∞ that is the usual convolution product. As it is easily seen the support supp(f ? g) of the element f ? g can be grater of the intersection supp(f ) ∩ supp(g): this means that the product is not local. The enveloping algebra of Hol(F). it does not increase the support of the two functions. These remarks allow to state that by considering several copies of F we could define Multi-differential operators by means of the action on functions (D1 ⊗ D2 ⊗ · · · ⊗ Dn )(f1 . it does not says that every commutative C ∗ -Algebra is the algebra C 0 (M). say multi-derivation maps X1 ⊗ X2 ⊗ · · · ⊗ Xn Out of these multi-operators it is possible to define higher order differential opera- tors when an ordering is defined. This means that the identification between vector fields and derivations of the algebra of functions cannot be based on the commutativity alone. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 18 Contents i. the holomorph of the Abelian Lie algebra F. fn ) = D1 f1 · D2 f2 · · · Dn fn It is also possible to replace first order differential operators with derivations. For instance one could write D1 (D2 (· · · Dn f )) or X1 (X2 · · · Xn f ). indeed. Remark 2. we have ordering problems. It is worth spending at this point some more words on the duality between vector fields and derivations of the Abelian algebra F(M).. we consider the subspace of square integrable functions on M as an Abelian vector group.. · · · . May 9. Any self-adjoint operator will give rise to a one-parameter group of transformations which will never respect the algebra structure. i.e.e. are vector fields if and only if the product of two functions is local. unless it is described by a first order homogeneous differential operator.e. The multiplication by p is a derivation of this product as shown in the following: Z +∞ (p(f ? g))(p) = p f (p − q)g(q)dq = −∞ Z +∞ ((p − q)f (p − q)g(q) + f (p − q)qg(q))dq = ((pf ) ? g)(p) + (f ? (pg))(p) −∞ . The differential operator we obtain in this manner suffers of the same problems we have when defining a “quantization procedure” for polynomial functions in the momenta.

D2 ) = D1 · D2 + D2 · D1 The first bracket defines a Lie algebra structure. [D2 . D1 ] . the algebra of differential operators may be considered an associative algebra with bilinear operation given by the composition (D1 · D2 )(f ) = D1 (D2 f ) . D3 )) − ((D1 . (D2 . D3 ]] − [[D1 . It contains the subalgebra of zeroth order operators (multiplication operators). D2 ) + (D1 . we map F(R) with the convolution product. D2 ] + [D1 . We should point out that there could be local products which are associative and commutative. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 19 Clearly this is not a vector field. indeed. This is the case for f1 ·K f2 = f1 · K · f2 with K any function on the manifold. Clearly. [D. D1 ] . whose derivations are vector fields. [D. D2 ]] and [D. There- ∗ ∗ fore this shows that the same commutative C -Algebra (the C -Algebra of mul- tiplicative operator on the Hilbert space L2 (R)) is represented as the pointwise algebra of continuous functions on R only in the second case. D2 )] = ([D. (D1 . D2 ]) It is quite remarkable that we have the identity (D1 . but they do not generate the tangent space at each point of the manifold. if dim(M ) = n. that φ((f ? g)(p)) = φ(f )(x) · φ(g)(x) ∂ and the derivations of this product are the usual vector fields X = A(x) ∂x . It is well known. Of course we have the Jacobi identity [D. D2 ] . [D1 . in particular. D2 ). However by means of Fourier transform φ : F(R) → F(R). D3 ) = [D1 . Out of this binary operation it is possible to produce another bilinear binary operation. It may be compared with the space of all linear endomorphism End(F). May 9. the second one a Jordan algebra structure. namely the commutator [D1 . D3 ] . derivations of the deformed product will be all the vector fields which admit K as a constant of the motion.1. into the algebra F(R) equipped with the usual pointwise product ·.at point of the manifold they will generate a vector space of dimensions n − 1. D2 ]] = [[D. 3. Brackets Following [14]. Remark 3. D2 ] = D1 · D2 − D2 · D1 and the anticommutator (D1 .

the commutator bracket van- ishes therefore the Poisson structure is an extra structure made out of outer deriva- tions (instead of inner derivations). which. Λ] = 2Λ ∧ Γ . 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 20 Contents i.16]). say {σ(D1 ). can be translated in the requirement that [Λ. g} = Λ(df.Γ) associated with a pair (Λ. In general a Poisson structure is associated with a bivector field Λ. · · · . dg) = Λjk ∧ ∂xj ∂xk The Jacobi identity requires X ∂Λkl lj jk  ik ∂Λ il ∂Λ Λij + Λ + Λ =0 i ∂xi ∂xi ∂xi Remark 4. It must satisfy the Jacobi identity. For any differential operator D of order k on M we define the principal symbol to be the multilinear map h h i i σ(D)(df1 . LΓ Λ = 0 . where Λ is a bivector field. dg) with 1 jk ∂ ∂ Λ= Λ j ∧ 2 ∂x ∂xk and ∂f ∂g Λ(df. By using the Lie algebra struc- ture of the algebra of differential operators we find that σ(D) is a symmetric k- multilinear map which defines a k-polynomial function along the fibers of T ∗ M. fˆ1 . D2 ]) Example 5. Clearly when the associative algebra is commutative. May 9. by using the canonical Poisson bracket on T ∗ M it is possible to define a bracket on symbols. According to Dirac. and Γ is a vector field. df2 . In this manner. · · · fˆk where fˆ is the zeroth order multiplication operator. D1 ] · D2 + D1 · [D. it would be a q-Poisson bracket defining a q-Poisson algebra ([15. σ(D2 )} = σ([D1 . say {f. fˆ2 .. D1 · D2 ] = [D. dfk ) = · · · D.e. in this case. A Jacobi bracket on the space F(M) of functions on a manifold M is a bracket {· . ·}(Λ. D2 ] is satisfied is called a Poisson bracket and the triple is called a Poisson algebra. they define a Lie-Jordan algebra. Γ). We recall the definition: A Lie bracket on an associative algebra such that the Leibnitz rule [D.

2. iΓ (θ ∧ (dθ)n ) = (dθ)n .Γ) = Λ(df. Then. g}(Λ. Therein it is possible to establish a one-to-one .Γ) = {f. gh}(Λ. its Clifford algebra has dimension 2n which is the same dimension as the exterior algebras Λ(V ) and Λ(V ∗ ). Clifford Algebras Given a vector space V. h}(Λ. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 21 so that the bracket may be written in the form {f.Γ) h + g {f. ·}(Λ. the coordinate expression of the Jacobi bracket reads     ∂f ∂g ∂f ∂g ∂f ∂g ∂f ∂g {f. one can consider the two sided ideal Ig (V ) generated by elements of the form v ⊗ v − g(v. writing the contact structure θ as: θ = dt − pa dq a . Indeed we have {f. g}(Λ. one can consider its tensor algebra ⊗k T (V ) = ⊕k=∞ k=0 V If the vector space V is equipped with a bilinear symmetric form g. g) that we will call ∨-product. g) Ig (V ) is the Clifford algebra associated to the space (V. For instance. g). v)1 The quotient space T (V ) = Cl(V. 1}(Λ. 3. The tensor product on T (V ) induces a product on Cl(V. on a contact manifold (M. g}(Λ. May 9. we can say that the Jacobi bracket is the coun- terpart of contact structures just as the Poisson bracket is the counterpart of the symplectic structure for symplectic manifolds. The Jacobi bracket turns out to be relevant for the geometrical formulation of thermodynamics ([17]) and for the covariant description of relativistic point particles. θ) we can define Λ and Γ by means of the expressions: iΛ (θ ∧ (dθ)n ) = n θ ∧ (dθ)n−1 .Γ) gh . If the vector space has di- mension n.Γ) − {f.Γ) acts on C ∞ (M ) as a first order differential operator not as a derivation. This bracket does not satisfy the Leibnitz rule and {f. ∂pa ∂q a ∂q a ∂pa ∂t ∂pa ∂pa ∂t In the case of contact manifolds. dg) + (LΓ f )g − f LΓ g . where dim(M) = n + 1.Γ) = − + g − p a − f − pa .

which will play the role of the space of spinors[19]. On the other side exterior derivative d is not a derivation of this new product. the n − k-form given by ∗α(v1 . May 9. ia1 = iea1 is the contraction with {ea }. and the sections Λ(M) = Γ(A(M)) form a module on this ring. we will denote its dual basis by {ea }. vN −k ) = αa1 ···ak g a1 b1 · · · g ak bk ω(eb1 ∧ · · · ∧ ebk ∧ v1 ∧ · · · ∧ vn−k ) By using ∗-operator one can build another homology operator which decreases the degree of the antisymmetric tensor on which acts. · · · .3. Exploiting this correspondence we realize a Clifford algebra on Λ(V ∗ ) and Λ(V ) by defining the following product X s(s−1) φ∨ω = (−1) 2 g a1 b1 · · · g as bs (γ s (ia1 · · · ias φ) ∧ (ib1 · · · ibs ω)) s where φ ∈ Λ (V ) and ω ∈ Λp (V ∗ ). 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 22 Contents correspondence between these vector spaces ([18]). previous constructions can be extended to this setting and we can analyze the behaviour of the algebraic differential calculus with respect to a Clifford algebra structure. Let us consider an orthonormal basis {ea } of the vector space V. we construct the exterior algebra A(M) over M which is ∗ A(M) = ∪m Λ(Tm M) Practically this means that. Then it is possible to define an antisymmetric tensor of maximal degree on V as ω = e1 ∧ · · · ∧ en which we will call a volume form. the field K of coefficients of the vector space V is replaced by the ring of functions F(M). 3. whereas the Lie derivative Lv is a derivation iff the vector v is in the centre of the Lie algebra V. we are able to define a possible Dirac operator as D = (d + δ) It will act on the exterior algebra. If V has an algebra structure as well. Clifford Modules By using the same ideology of gluing together vector spaces at each point of a manifold M. It is possible to show that iv is still a graded derivation of the ∨-product with degree −1. as already explained. say δ : Λk (V ∗ ) → Λk−1 (V ∗ ) δα = (−1)n−k (−1)sign(g) ∗ d ∗ α Having extended our algebraic calculus. If the manifold is equipped with a . defining k ∗ an orthonormal basis of the vector space V. This choice allows us to define an operator ∗ : Λk (V ∗ ) → Λn−k (V ∗ ) which associates to the k-form α = αa1 ···ak ea1 ∧ · · · ∧ eak .

we can introduce on Λ(M) a F(M)-valued scalar operator (·. and this construction can be also described in the algebraic setting of Lie Algebroids ([21]). βi = (α. β) = ∗(α ∧ ∗β) and saying that it is zero if the two forms have different degrees. ·) : Λk (M) × Λk (M) → F(M) (α. Let us come back to Λ(M). If g is a nondegenerate symmetric bilinear form it is associated to a metric tensor. Remark 6. As this product does not increase the support of the sections. say δ : Λk (M ) → Λk−1 (M ) δα = (−1)n−k (−1)sign(g) ∗ d ∗ α In terms of d and δ. Once again the contraction iX remains a derivation of degree −1. ∗ is the Hodge dual operator which associates to each k-form α = αa1 ···ak θa1 ∧· · ·∧θak . a covariant derivative is defined. May 9. which we will call g. the Lie derivative LX is a derivation if X is a Killing vector field. this derivations are associated to vector fields. If one considers the Levi-Civita connection on the tangent bundle TM. As before we can analyze the behaviour of the exterior differential calculus with re- spect to this Clifford structure. If we want a scalar product we can define Z hα. β)Ω (51) M The next step is the introduction of another boundary operator. where {θa } is an orthonormal basis of the module Λ1 (M ). It is related to the choice of the volume form Ω Ω = θ1 ∧ θ2 ∧ · · · ∧ θn . the exterior derivative d is not a derivation. the Laplace-Beltrami operator ∆ can be written as ∆ = dδ + δd = (d + δ)2 Therefore a Dirac-like operator is given by the scalar differential operator . Let us consider a bilinear form which is also non- degenerate. It can be shown that the covariant derivative ∇X is a derivation of the ∨-product. This means that Λ(M) is a Clifford module ([20]). By exploiting the scalar product defined at each point. the (n−k)-form 1 ···ak ∗α = abk+1 ···bn αa1 ···ak θ bk+1 ∧ · · · ∧ θbn . 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 23 bilinear symmetric form we can define a ∨-product at each point and the resulting bundle A(M) will have also a Clifford algebra representations.

i. Remark 7. Interior product is a derivation of degree −1. and the Lie derivative LX . then δ is not the adjoint of d and the two operators are related by Z Z hdα . i.. ·). X1 . whith β ∈ Λk−1 (M). forms α ∈ Λk (M) such that α = δβ with β ∈ Λk+1 (M). the interior product (or contraction) iX . if M has a (smooth) boundary ∂M . Here. • the space of coexact forms. The main change appears in the definition of a suitable ∗-operator and consequently of a δ.. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 24 Contents D = (d + δ) (52) as it is a square-root of the Laplacian.. . Xk−1 ) := α(X .e. Remark 8. that is. “every vector field X can be written as the sum of a gradient of something plus the curl of something else”. May 9. Let us note two things: • First of all one can notice that if M has no boundary δ is the adjoint of d with respect to the scalar product (51). Xk−1 ) . δβi = d (α ∧ ?β) = i∗ (α ∧ ?β) M ∂M where i : ∂M → M denotes the immersion of ∂M in M . and its action is defined as follows iX : Λk (M) → Λk−1 (M) (53) α 7→ iX α (54) iX α(X1 . ·). an useful theorem is the so-called Hodge decomposition theorem [5].. on R3 .. a gradient vector field Y is defined as G(df . where f is a function and G is the inverse of a metric tensor g. differential forms α such that ∆α = 0. forms α ∈ Λk (M) which can be written as dβ. and a curl vector field Z is defined as G(δβ . • the space of harmonic forms. . There are only other two derivations of the wedge product which are covariant with respect to this group. (55) . . The expression in equation (52) for a Dirac-like operator is scalar with respect to the diffeomorphisms group. A consequence of this theorem is the well known fact that.e. that is. on the contrary... • The same procedure can be applied also if g has a kernel. In the context of differential equations written on the space of differ- ential forms. βi − hα . where X is a generic vector field. It states that the Hilbert space of differential forms can be decomposed into the direct sum of three orthogonal subspaces which are respectively: • the space of exact differential forms. where β is a differential two-form.

(57) According to the scalar product on the space of differential forms. when α. when it acts on a k-form. we use only LX + 21 div(X) to deal with a symmetric differential operator. (59) ∂M where the right-hand side appears only in the case of manifolds with a boundary. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 25 The Lie derivative is a derivation of degree 0. Their idea is to define a differential calculus intrinsically discrete which mimics the usual differential calculus on continuous manifolds but it is not derived by means of approximation procedures. and its action on differential forms can be characterized by means of the Cartan formula: LX : Λk (M) → Λk (M) (56) LX = iX d + d iX . [d . Example 9. ·) ∧ β) (58) where α ∈ Λ1 (M). LY ] = i[X . In particular. on functions. Thus.Y ] . L†X β ) = iX (α ∧ ∗β) . The three graded derivations d . At this purpose let us start with a manifold made up of a set of points and consider the complex K (see for instance [24]) of dimension n generated by linking them one to each other. It is well known ([22]) that the adjoint of iX is i†X = (−1)n(k+1)+1 ∗ iX ∗. and G is the inverse of the metric tensor g. β are functions. (iX α. i†X β) = (α. β ) − ( α . (60) where Ω is the volume form associated with the metric tensor g. An interesting application of the concept of duality in order to define a differential calculus can be found in the discrete exterior differential calculus built by Desbrun et al in [23]. such as general relativity. LX ] = 0 . May 9. (61) The study of these differential operators is fundamental when one deals with theories that are covariant with respect to the diffeomorphisms group. Since we are interested in a Dirac-Kahler operator in this discrete setting. d] = LX . G(X. the adjoint of LX is L†X = (−1)k(n−k)+1 ∗ LX ∗ and it satisfies the following relation Z ( LX α . In particular. the adjoint L†X is the first order differential operator LX + div(X). with div(X) a scalar function defined by: LX Ω = div(X) Ω . iX . β) = (α. β is a function. these op- erators are not symmetric. LX close a graded Lie algebra according to the following rules: [iX . let us introduce a . [iX . On the other side.

therefore. every open subset of a manifold can be identified with its tangent space. This dual complex will play a fundamental role in the definition of a Hodge operator. vk ). vτ (0) · · · vτ (k)   l   β . The pairing between a k-chain σ k and a k-cochains αk will be denoted by αk (σ k ) = αk . that is the complex which is made up of circumcentres (or barycentres) of each n-simplex of the complex. vk ) = 0 ⇒ vj = vk • Simmetry : d(vj . σ k . Giving a local metric. vk ) + d(vk . vl ) In other words we are considering any single n-simplex as an open subset of the discrete manifold: as already said. Z) to the additive group R is the space of cochains and will be denoted as Ωk (K). May 9. vk ) = d(vk . σ k+l = sign(τ ) (k + l)! |σ k+l | τ ∈Sk+l+1 αk . vk ) ≥ 0. This local metric is a map d : {(vj . Z) the space of k-chains. It can be thought as the result of integrating a differential forms along a path. Coming back to the space of cochains. indeed. vj ) = 0 • Strictly Positive : d(vj . vτ (k+1) · · · vτ (k+l) (63) where |σ k | denotes the volume of the k-simplex calculated according to the local distance and the local embedding in Rn . The space of Homomorphism from Ck (K. it can be equipped with the wedge product ∧ : Ωkd (K) × Ωld (K) → Ωdk+l (K) defined by k 1 X |σ k+l ∩ ?vτ (k) | α ∧ β l . In this discrete setting the space of cochains replaces the space of differential forms. vj ) • Triangle inequality : d(vj . 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 26 Contents local metric on this complex by defining a distance map among vertices of each n- simplex in the complex. j≥k   where c(σ j ) is the circumcentre of the simplex σ j and c(σ k ) · · · c(σ j ) · · · c(σ n )) denotes the convex hull of the points c(σ j ) with j ≥ k and εσk is a factor related with the orientation of the dual simplex. Let us call Ck (K. vl ) ≤ d(vj . ∀ [vj vk ] ∈ σ n } → R must satisfy some properties: • Positive : d(vj . is equivalent to the assignment of a scalar product on the tangent space at each point. By means of this dual subdivision one can associate to each k-simplex a (n−k)-simplex of the dual complex X ? (σ k ) = εσk c(σ k ) · · · c(σ j ) · · · c(σ n ))   (62) σ j . . that is the free abelian group of formal finite sums of k-simplices in K with coefficients in Z. and d(vj . A further ingredient is the so called dual complex.

respectively. May 9. Since there exists a dual complex one can define the discrete Hodge Star operator as the map ∗ : Ωkd (K) → Ωn−k d (K) which acts as follows 1 k k 1 ∗α . we can explicitly write and try to solve Dirac-Kahler equation for a one-dimensional discrete manifold. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 27 The boundary operator ∂k : Ck (K. It is possible to show that exterior differential d : Ωd (K) → Ωd (K) is a deriva- tion of the wedge product but this product. vj+1 ) = d(vk . v̂j . vk ] k X ∂k σ k = (−1)j v0 . vj+1 ]. · · · v k is the simplex obtained by omitting the vertex vj . Z) is a homomorphism de- fined by its action on the simplex σ k = [v0 . · · · . let us consider a local metric which is the same at each point. v̂j . Therefore one can write Dirac-Kahler operator acting on a generic cochain as: X  (1) (1) (0) (0) (0) (1)  i(d − δ)( α(0) + α(1) ) = i (αj − αj−1 )ej + (αj+1 − αj )ej (68) j We can diagonalize this operator by choosing (0) αj = Ae−ijlk (0) αj = Be−ijlk If one replaces this ansatz in the eigenvalue equation i(d − δ)( α(0) + α(1) ) = λ( α(0) + α(1) ) .σ = k (66) |σ | |?σ | Now the codifferential operator is δ : Ωkd (K) → Ωk−1 d (K) δβ k = (−1)nk+1 ∗ d ∗ β k (67) Having at our disposal all the necessary ingredients. The exterior derivative dk : Ωkd (K) → Ωk+1 d (K) is defined as the coboundary operator dk αk . σ k+1 = αk . one can write 0-cochains and 1-cochains as P (0) (0) α(0) = j αj ej P (1) (1) α(1) = j αj ej . ∂k+1 σ k+1 . (65) This expression makes Stoke’s theorem true by definition. however if we restrict to the subset of closed cochains the product is associative. · · · v k . vk+1 ) = l. For simplicity. ?σ k k α . In particular let us consider a complex made up of the union of an infinite number of one-dimensional simplices [vj . is no more associative. · · · . Z). that is d(vj . → Ck−1 (K. If we denote (0) (1) the dual elements of the simplices vj and σj = [vj vj+1 ] by ej and ej . in general.   (64) j=0   where v0 . · · · .

the states of classical or quantum systems in the C ∗ -algebraic approach are the normalized.. 4. May 9. For simplicity. However one can generalize the expression (68) to the case of non constant metric and eventually we get the result that local metric behaves like a coupling constant for this system. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 28 Contents one gets the following pair of equation for the coefficients A and B: iA(1 − eikl ) − λB = 0 −λA + iB(e−ikl − 1) = 0 This system admits a non vanishing solution if and only if the determinant of the coefficient matrix is zero. when A is commutative with dimension N < +∞. Manifolds with boundaries and corners In accordance with the C ∗ -algebraic approach to classical and quantum theories. namely. In the finite-dimensional classical case. It is possible to use this calculus to deal with gauge theories on lattices. . that is 2 λ2 + e−ikl 1 − eikl = 0 . let us recall that if |ψi ∈ CN and j |ej ihej | = I is an orthonormal resolution of the P identity.. the study of such a model needs a deeper investigation. the manifolds we shall be interested in arise as convex subsets of linear spaces. has been obtained by considering a constant metric. positive. p2 . Clearly they form a manifold with corners arising P from the Cartesian product of N copies of R+ intersected with the affine space j pj = 1. pN ) in RN with the properties pj ≥ 0 and P j pj = 1. However.. (69) hψ|ψi hψ|ψi The Euclidean structure of CN defines a distance on the space of probability vectors which is associated with the metric: . real linear functionals on the C ∗ -algebra A of the system. There is a simple way to “quantize” a classical probability vector. we shall consider finite-dimensional systems. which agrees with the results provided by other models describing fermions on a lattice. we have a probability vector associated with any |ψi and any partition by setting: tr (|ψihψ|ej ihej |) hψ|ej ihej |ψi pj := = . whose solutions are   p kl λ=± 2 (1 − cos(kl)) = ±2| sin | 2 This simple expressions. indeed. which falls outside the purpose of this paper and will be presented somewhere else. the states of the system can be identified with the probability vectors. the vectors p~ ≡ (p1 . Specifically. that is.

given x. (71) 2 where h ip~ denotes the expectation value with respect to the probability vector p~. pN thus obtaining a “polar” representation of a generic mixed state associated with CN . the differential of the xj are dxj = √ . 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 29 N +1 X 1 X dpj dpk ds2 = gjk dpj dpk = δjk . nor a differential manifold with smooth boundary. May 9. P and hd ln p~i is actually zero because of the constraint j pj = 1. the 2 pj Fisher-Rao metric in appropriate coordinates reads ds2 = j dxj dxj (see [2]). (70) 4 j=1 pj j. However. and it can be shown ([6.. we would get the positive hyperoctant P P of S N −1 . if we consider z j = eıϕj pj we would go back to the quantum picture where z j = hej |ψi. In this picture.k p This is the so-called Fisher-Rao metric. the flat simplex for N = 3 becomes the round octant of dpj the sphere. Specifically. DA is a subset of the dual u∗ (N ) of the Lie algebra u(N ) of the unitary group U (N ).e. . all the probability vectors with the same number of non-zero elements would be on a given facef (corner) if they occupy the same f Recallthat a face of a convex set K is a convex subset F ⊂ K such that. Furthermore. In setting xj := pj we have decided to consider p the square root within the real numbers. At the “classical” level. P p Remark 10. and thus. This is the pullback to the Hilbert space H of the canonical Hermitean tensor on the complex projective space P(H) ([25]). The space DA of quantum states is not a differential manifold. if αx + (1 − α)y ∈ F then x. states becomes normalized. i.26]) that it k is the union of different strata DA labelled by the rank of the quantum states they contain. we would reconstruct a vector in CN +1 and an Hermitean tensor: 1 h= [hd(ln p~) ⊗ d(ln p~)ip~ − hd(ln p~)ip~ ⊗ hd(ln p~)ip~ ] + 4 i + hd~ ϕ ⊗ d~ ϕip~ − hd~ ϕi ⊗ hd~ ϕip~ + [hd (ln p~) ∧ d~ ϕip~ − hd (ln p~)ip~ ∧ hd~ ϕip~ ] . y ∈ F .. We notice that if we set xj := pj . A different way to “quantize” classical probability vectors would be to consider a probability vector as a diagonal matrix in the C ∗ -algebra A = B(H) and act on it with elements of the unitary group U (N ):   p1 ρ = U†   . positive Hermitean matrices acting on CN . For example.  (72) . we get 1 = j pj = j x21 + x22 + · · · + x2N . y ∈ K.  U. hence.

Thus. In what follows we are going to consider the geometry of DA according to [6. the space of quantum states DA . when a is looked at as an element A of A. and the algebraic ∗ properties of the space of observables OA of which the ambient space OA is the dual space.2] for a detailed discussion.26. We will describe the geometry ∗ of DA exploiting the ambient space OA . Then. This shift from U (N ) to SU (N ) comes from the coniugacy action on states. The linear space OA is not an associative subalgebra of the C ∗ -algebra A. The geometry of OA and DA Our stratified manifold. (73) 2 and a Jordan algebra structure: g This condition is equivalent to a = b2 for some b ∈ O . it carries a Lie algebra structure: ı [[a . and if we allow for a nonlinear action of the complexification of the unitary group. At the quantum level. In this case.27. OA becomes isomorphic to the set of self-adjoint linear operators on H. with respect to a non-linear action. each stratum is an orbit. or.26. we will argue that a notion of vector fields (derivations) transversal to the strata is needed in order to describe physically interesting processes such as decoherence. We shall consider only the case in which A is a finite-dimensional simple C ∗ -algebra with unity. an element ρ ∈ OA is a quantum states if ρ(a) ≥ 0 for every a ∈ OA with positive semi-definite spectrumg . we can map any state of a given rank into any other state of the same rank. denoting with I the identity element ∗ in A. in which case it is necessarily isomorphic to the set B(H) of bounded linear operators on a complex Hilbert space 2 H of dimension (dim(H)) = dim(A) = N (see [28]). it is equivalent to the existence of b ∈ A such that a = bb† .27. which is a linear space. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 30 Contents position in the column matrix describing a state.2]. May 9. considering the geometrization of the dynamics of open quantum systems.1. However. for which the rank of the states is not preserved. the elements of the linear space OA are identified with observ- ables. As said before. and if ρ(I) = 1. and the space DA of quantum states is a convex subset of the dual space ∗ OA of the space of observables. of the complexification of SU (N ). Specifically. We shall limit ourselves ∗ to a quick revision of the geometry of OA and DA . b ∈ OA ⊂ A . ∗ 4. probability vectors with the same elements can be mapped one into the other by means of the unitary group. The elements of OA are identified with the observable of the theory. . we refer to [6. is a convex subset of ∗ the set OA which is the dual space of the real vector space OA defined by the real elements of a C ∗ -algebra with unity A. b]] = − (ab − ba) with a.

2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 31 1 a. May 9.

b= (ab + ba) (a .

b) .

a2 = a .

(b .

This means that OA is an algebra with respect to both the products. (74) 2 making up what is known as a Lie-Jordan algebra structure on OA ([15. b ∈ OA ⊂ A . and it is such that [[ . ]] is a derivation for . a2 ) with a.16]).

: [[a .

c]] = a . b .

c]] . c]] + [[a . [[b .

b . (75) and it is such that the deviation from associativity of the Jordan product is pro- portional to the deviation from associativity of the Lie product: a.




c]] ]] − [[ [[a. c ∈ OA ⊂ A .c = λ2 ([[a. the associative product is recovered by setting ab = a . b. [[b. b]] . (76) The constant λ is there to remind us that if we want to recover the associative product of the C ∗ -algebra A we have to fine-tune the Jordan and Lie product. c]]) with a. specifically.

being OA and OA finite-dimensional vector spaces. When the appropriate physical units are introduced. + . ·) where the product · is the pointwise product (f · g)(ξ) := f (ξ)g(ξ). b]]. Therefore. This is possible because. associated with a basis {aj } in OA there is the coordinate system {xj } on OA ∗ given by xj (ξ) := faj (ξ). g ∈ F(OA ) and ξ ∈ OA . This means ∗ that. ∗ ∗ ∗ with f. b + ı [[a. the product of two linear functions is ∗ a quadratic function. we can define ∗ ∗ a linear function fa : OA → R given by fa (ξ) := ξ(a) with ξ ∈ OA . However. Fl (OA ) is not closed with respect to the pointwise product in F(OA ). It is immediate to see that fa+b = fa + fb . given any element a ∈ OA . and thus the linear space OA ∗ ∗ is represented as a linear subspace Fl (OA ) of the linear space F(OA ). This can be done defining the following products: (fa ∗ fb ) (ξ) = f[[a . it turns out that λ is propor- tional to Planck constant ~.b]] (ξ) . (77) (fa . we have to define two new products on Fl (OA ) so that it becomes a Lie-Jordan algebra. ∗ We will now represent OA and its Lie-Jordan structure in the space F(OA ) of ∗ functions on OA . the differentials dxj of the coordinate functions form a basis of the cotangent space Tξ∗ OA ∗ at each point ξ ∈ OA∗ . The lin- ∗ ∗ ear space F(OA ) of smooth functions on OA has a natural structure of algebra ∗ (F(OA ) . indeed. furthermore. and.

fb ) (ξ) = fa.

(78) As it is clear. and . ∗ represents the Lie product of OA .b (ξ) .

and it is ∗ a matter of straightforward calculations to show that. the Jordan product. they make Fl (OA ) . together.

This means that the unary operation Lfa (fb ) := fa ∗ fb ∗ is a derivation of . 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 32 Contents into a Lie-Jordan algebra. May 9.

Now. for all fa . ∗ we can use ∗ and . fb ∈ Fl (OA ). since linear functions generate the cotangent space Tξ∗ OA ∗ ∗ at each ξ ∈ OA .

dfb )(ξ) := (fa . to define two contravariant tensor fields on OA : Λ(dfa . (79) R(dfa . dfb )(ξ) := (fa ∗ fb ) (ξ) = f[[a.b]] (ξ) .

fb ) (ξ) = fa.

where {aj } is a basis in OA .b]] a linear function. a b (83) h i Xef . fb . A direct calculation shows that. the space of observables OA is the space of Hermitean matrices on H. being {fa . Hamiltonian and gradient vector fields satisfy the following commutation relations:: h i Yefa . ·) 1 X Yef := R (df . ·) . when f. This means that we can identify an element ∗ ξ ∈ OA with an Hermitean matrix ξe ∈ OA . furthermore. Being A ∼= B(H).f } . given f ∈ F(OA ). (81) ∂xj ∂x ∂x ∂x These two tensor fields allow to define the notions of Hamiltonian and gradient ∗ ∗ vector fields on OA . a b (84) h i Xef . Yefb = −X e{f .f } . and (82) λ where λ is the parameter appearing in formula (76). the generalized distribution: n o ∗ ef or Z = Yef with fa ∈ Fl (O∗ ) Dgl := Z ∈ X(OA ) : Z=X a a A (86) is integrable. With the help of this isomorphism and . Specifically.f } . being in finite-dimension. g are linear functions say fa . fb } = fa ∗ fb = f[[a . (80) ∗ Note that Λ and R are tensor fields on the whole OA . we define the Hamiltonian vector field Xf and the gradient vector field Yf as follows: e e ef := Λ (df . a b a b (85) Consequently. Selecting the coordinate ∗ system {xj } on OA given by xj (ξ) := faj (ξ). X a ef = X b e{f .b (ξ) . every choice of basis in OA determines ∗ ∼ a non-canonical isomorphism OA = OA . we obtain the coordinates expression of Λ and R: ∂ ∂ ∂ ∂ Λ = cljk xl ∧ and R = dljk xl ⊗ . Yef = Ye{f .

both counted with multiplicities. the rank of ξ ∈ PA . theorem 11 applies. k− ) := {ξ ∈ OA : k+ . The family: ∗ ∗ OA (k+ . connected submanifold ∗ ∗ of OA . we see that the leaves of the foliation generated by Dgl are the orbits ∗ e † on OA . with n = dim(H). and its dimension is (2nk − k 2 ) with n = dim(H). Accordingly. k− ) at ξ is identified with the linear subspace of OA the elements B of which satisfy the condition: hBx|yi = 0 ∀x.e and k− for the number of negative eigenvalues. Specifically. Define the rank rank(ξ) of ξ ∈ OA as the matrix rank of its representative ξe ∈ OA . The ∗ ∼ representative ξe ∈ OA of ξ ∈ PA through the isomorphism OA = OA introduced above is a positive semi-defined Hermitean operator on H. May 9. every OA (k+ . Denoting with P k the set of elements of eigenvalues of ξ. we write k+ for the number of positive eigenvalues of ξ. Furthermore. k The set PA is the disjoint union of the smooth manifolds PA for 0 ≤ k ≤ n2 . ξe = AA† for some A ∈ B(H). and the leaves of the associated foliation are the orbit of the U(H) action induced by ξe 7→ UξU e †. rank(ξ) = k+ + k− . A k PA of rank k. This set is the cone of positive hermitean functionals on A. Furthermore. it is OA (k. it is clear that the GL(H) action introduced above preserves PA since: † T ξeT† = T AA† T† = TA (TA) . let use denote with PA the set of ξ ∈ OA such that ξ(a2 ) ≥ 0 for a ∈ OA . specifically. y ∈ Ker( ξe) . The main result in [6] is the following: Theorem 11. . and thus PA is a smooth. connected submanifold ∗ ∗ of OA . Note of the GL(H) action on OA induced by the GL(H) action ξe 7→ TξT that the distribution DΛ consisting of Hamiltonian vector fields associated to linear functions is itself integrable. for every smooth curve γ : R → ∗ k k OA such that γ(τ ) ∈ PA for all τ . is an open submanifold of OA . 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 33 the work [6]. k− ≥ 0 . k− ) is a smooth. ∗ At this purpouse. The set of all such operators is denoted as P(H). an important result in [6] states that. ∗ The leaves of Fgl have different dimensions. If rank(ξ) = k. hence. Clearly. this action preserves the rank and the number of positive and negative e that is. γ(τ ) ∈ PA implies γ̇(τ ) ∈ Tγ(τ ) PA . k = k+ + k− ≤ n} (87) ∗ ∗ of subsets of OA is exactly the family of orbits of the smooth GL(H) action on OA ∗ introduced above. n ∗ Note that the top stratum PA . (88) With the aid of theorem 11 it is possible to show that DA is the disjoint union k ∗ of smooth manifolds DA ⊂ OA . The tangent space Tξ OA (k+ . 0).

The k k k function fI is regular on each PA . since eI = 1. At this purpouse. However. DA is connected.30]). we start considering the so-called expectation value functions ea (ξ) = ffaI (ξ) (ξ) on it. we will follow a different path. in order to correctly define tensors on O. (89) ∗ Again. only on ∗ the open submanifold O of OA composed by elements ξ ∈ OA such that ξ(I) 6= 0. The dimension of DA 2 k is (2nk − k − 1). (90) ΛDA (dea . DA is a regular submanifold of PA . Now. and it holds PA = DA × R . Now we want to find contravariant tensor fields ΛDA and RDA such that k the Hamiltonian and gradient vector fields associated with them are tangent to DA for all k. This results will bring important physical consequences when dealing with the dynamical evolution of a quantum system generated by a Kossakowski-Lindblad operator ([29. in general. A natural way to proceed would be to use Dirac’s prescription for Poisson brackets with constraints. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 34 Contents Recall that a quantum state ρ is an element in PA such that fI (ρ) = 1. RDA (dea .b]] (ρ) . we define the tensor fields ΛDA and RDA requiring that: ΛDA (dea . deb )(ρ) = e[[a. Topo- k k ∼ k + k logically. A direct calculation shows that the Hamiltonian vector field Xef associated with the function f by means of Λ is tangent to the affine subspace fI (ξ) = 1 for all functions f . ∗ Note that these functions are not defined on the whole OA but. therefore. and the tangent space Tρ DA at ρ is identified with the linear subspace of u∗ (H) consisting of traceless elements B satisfying the condition: hBx|yi = 0 ∀x. May 9. However. deb )(ρ) = ea. gradient vector fields do not preserve the affine subspace fI (ξ) = 1. however. dfI )(ρ) = 0 ∀a ∈ OA . and the affine subspace fI (ξ) = 1 can be identified with the quotient of O with respect to the action of the dilations. hence. the tangent vector γ̇(τ ) of every smooth curve γ : R → OA lying entirely in k DA is tangent to the stratum DA to which γ(τ ) actually belongs. the expectation value functions are not enough to generate the cotangent bundle of O. we have to consider the function fI . y ∈ Ker( ρe ) . We choose the expectation value functions because these functions are invariant with respect to dilations.

(91) RDA (dea . dea )(ρ) = RDA (dfI . it is interesting to notice that RDA introduces a measure of nonclassicality of the state in connection with observables a and b. Note that the Jordan algebra structure of OA is no more represented by the symmet- ric tensor RDA . it measures how much the local pointwise product differs from the quantum nonlocal . however. ∀a. dfI )(ρ) = RDA (dfI . Indeed.b (ρ) − ea (ρ)eb (ρ) . dfI )(ρ) = 0 ∀a ∈ OA . b 6= I .

. With a little patience. i. it is possible compute the commutators of the “new” Hamiltonian. eb }DA := ΛDA (dea .eb } . the Hamiltonian and gra- dient vector fields defined above provide an action of the complexification sl(N. we can think of it as defining a bidifferential operator on O.b]] . and see that the “new” Hamiltonian and gradient vector fields are tangent to the integral submanifolds of the distribution generated by the “old” Hamiltonian and gradient vector fields. “new” Hamiltonian and gradient vector k fields are tangent to DA for all k. (92) λ Direct calculations show that: [Yea . (98) Example 13 (Two-level quantum system). Specifically. (94) [Xea . eb )DA := RDA (dea . 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 35 product when evaluated on ρ. dea ) gives the covariance of the observables a and b on the state ρ. (96) while the Jordan product is represented by means of the following bracket: (ea . C) of the Lie algebra su(N.eb } . This means that the flow of the “new” Hamiltonian and gradient vector fields maps positive elements of rank k in O into positive elements of rank k in O. From the statistical point of view. Xeb ] = X{ea . a bivector plus a homogeneous part. However. ·) + ea · . Remark 12. for a quantum system with N levels. so that we obtain the first order differential operator: Da (·) = (ea . .eb } . )DA is not associated to a bivector on O. Yeb ] = Y{ea . C). ·) and Yf := RDA (df . are defined as: 1 Xf := ΛDA (df . (95) and thus we find that. or gradient. May 9. the Lie product is represented by means of the bracket associated to ΛDA : {ea . (97) The Jordan bracket ( . deb ) + ea eb . The “new” Hamiltonian and gradient vector fields. its action on the constant function eI is not zero. ·) .e. deb ) = e[[a. indeed. (93) [Xea . Yeb ] = −X{ea . RDA (dea . vector fields with the “old” Hamiltonian and gradient vector fields. that is. ·)DA = RDA (ea . We can represent the Lie-Jordan algebra OA using brackets on ex- pectation value functions ea on O.

DA is the 3-dimensional solid ball and the two strata are: • the pure states x21 + x22 + x23 = 1. x3 } on OA by means of which the repre- sentative ρ of a state is written as: 1 ρ= (σ0 + ~x · ~σ ) (100) 2 with x21 + x22 + x23 = 1. On the affine subspace x0 = 1. (102) ∂xj ∂xk j. In this case. σ2 . (103) ∂xj while Hamiltonian ones read: ∂ Xj = jkl xl . x2 . (104) ∂xk h As shown in [6] this is the only case in which D A is a differential manifold with a smooth boundary. x1 . It should be noticed that while pure states are represented by a compact man- ifold without boundary. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 36 Contents To illustrate our general arguments we consider the example of the two-level quantum system. and a ∗ Cartesian coordinate system {x0 . σ3 } there is an isomorphism OA = OA . Specifically. namely.k. the expressions for RDA and ΛDA are: 3 3 X ∂ ∂ X ∂ RDA = ⊗ − ∆DA ⊗ ∆DA with ∆DA = xj . h and it is a proper manifold with boundary . DA has only two strata. • the mixed states x21 + x22 + x23 < 1. The C ∗ -algebra A is isomorphic to B(H) for H ≡ C2 . Thus the space OA of the real elements is generated by the Hermitean (2 × 2) Pauli matrices:         10 01 0 −ı 1 0 σ0 = σ1 = σ2 = σ3 = . the stratum of mixed states is bounded but not compact. σ1 . . DA 1 2 and DA . May 9. (101) j=1 ∂x j ∂x j j=1 ∂xj 3 X ∂ ∂ ΛDA = jkl xl ∧ .l=1 Gradient vector fields associated with linear functions are: ∂ Yj = − xj ∆DA . and its closure is the whole of the ball. (99) 01 10 ı 0 0 −1 ∗ ∼ Associated to the basis {σ0 .

(108) 2 j=1 j=1 where ρ ∈ DA . hence. {Φτ } can “move” states across different strata. such that Φ0 is the identity transformation. ρ] − Vj† Vj . the Hamiltonian ones are tangent to the sphere of radius r for all r > 0: LXj r2 = jkl xl xk = 0 . We would like to describe such a dynamics in geometrical terms. a Markovian dynamics is described by the equation of motion: d ρ = L(ρ) ρ(t = 0) = ρ0 . (107) dt According to [29] and [30].30]. the expression for the linear generator L reads: 2 2 1 X−1 n r≤n o X−1 r≤n L(ρ) = −ı [H. and [31] chapter XXV). May 9. However. that is. Actually. for τ < 0 it fails to preserve positivity. Kossakowski-Lindblad vector fields The algebraic description of quantum systems associated to the C ∗ -algebraic for- mulation has led to an interesting formalization of the dynamics of open quantum systems in terms of the so-called Kossakowski-Lindblad generator L ([29. Furthermore. Furthermore. This particular feature makes {Φτ } a good candidate for the description of physical phenomena such a dechoerence. C). {Φτ } is well-defined and differentiable ∗ for all τ ∈ R on the whole OA . and { . Specifically. hence.2. (105) while the gradient ones are tangent only to the sphere of radius r = 1 (the pure states): LYj r2 = 2xj − xj (2r2 ) = 2(1 − r2 )xj =⇒ LYj r2 |r2 =1 = 0 . it maps states out of DA . and together they close the Lie algebra of SL(2. there is a linear vector field ΓL naturally associated with it. H† = H. it is possible that rank(Φ0 (ρ)) ≤ rank(Φτ (ρ)) when τ > 0. but. k the curve Φτ lies in some DA for all τ > 0. ∗ Recall that a differentiable curve γ : R → OA lying entirely in the full space of k quantum states DA must remain in the stratum DA it actually belongs to. Since L is a linear map. where the rank of the state is not preserved. ρ + Vj ρ Vj† . The integration of the equation of motion gives a one-parameter semigroup {Φτ } of completely-positive maps Φτ : DA → DA for τ ≥ 0. Vj are arbitrary elements of B(H). } denotes the anti-commutator. using vector fields. that is.  (106) 4. when {Φτ } is a semigroup. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 37 They are complete vector fields. we can write: .

both Hamiltonian k and gradient vector fields. May 9. (109) ] † P where K (a) is the action on OA of the adjoint map of K(ξ) = j VξV . and a Kraus vector field ZK . is defined requiring that ZK (fI ) = 0. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 38 Contents eH − YeV + ZeK . Moreover. in general. j=1 j Although very simple.a]] (ρ) + eG. Indeed. being tangent to DA . gradient vector fields are. the Kossakowski-Lindblad dynamics can be described using a fine-tuned combination Γ = XH + YG + ZK (110) of a Hamiltonian vector field XH with H ∈ OA . we would like to use the Hamiltonian and gradient vector fields associated with ΛDA and RDA in order to describe the Kossakowski- Lindblad generator L. are not able to move a state across different strata. while the generator L is linear. 2 and ZeK is the linear vector field associated with the linear map Pr≤n −1 Vj ρ V† . and V = j Vj† Vj . The action of a generic Γ on the expectation value functions reads: (Γ(ea )) (ρ) = e[[H . Accordigly. called Kraus vector field. It turns out that is possible to define a vector field ZK implementing the motion across strata. This vector field ZK . and requiring its action on expectation value functions to be: (ZK (ea )) (ρ) = eK ] (a) (ρ) − eV (ρ)ea (ρ) . YeV is Pr≤n2 −1 † the gradient vector field associated with V := 21 j=1 Vj Vj by means of R. both YeV and ZeK do not separately preserve the affine subspace fI (ξ) = 1 of which the space of quantum states DA is a subset. a gradient vector field YG with G ∈ OA . P Now. non-linear. ΓL = X where XeH is the Hamiltonian vector field associated with H by means of Λ. Indeed. However. this decomposition of the vector field ΓL is not adapted to the geometry of the space of quantum states. we can not use only Hamiltonian or gradient vector k fields tangent to DA to accomplish this task.

In conclusion. hence.a (ρ) + eK ] (a) (ρ) − eG (ρ) ea (ρ) − eV (ρ) ea (ρ) . (112) . (111) As it stands. choosing G = −V the resulting vector field is linear. Γ does not describe the Kossakowski-Lindblad dynamics since it is not linear. the the Kossakowski-Lindblad dynamics of a quantum system is described by the Kossakowski-Lindblad vector field: ΓL = XH + YG + ZK . The non-linearity of the vector field Γ is due to the last two terms.

however. (115) ∂x1 ∂x2 The integral curves of ΓL are radial lines lying in two-dimensional planes orthogonal to the x3 -axis. Example 14 (Phase damping of a qubit). if the initial state ρ is a mixed state. the Kossakowski-Lindblad P vector field ΓL reduces to the Kraus vector field ZK . ΓL generates a one-parameter semigroup of transformations. Conclusions In many physical situations. it will change the spectrum of ρ giving rise to dissipation. there is the need to develop methods to handle such special situations. Consequently. May 9. Let us consider the Kossakowski-Lindblad dynamics generated by equation (108) √ √ with H = 0. Indeed. Note that. with the notation of example 13. note that H = 0 implies XH = 0. giving rise again to a dissipative behaviour. The Kossakowski-Lindblad vector field on the affine subspace fI = 1 can easily be calculated. the flow of ΓL will take ρ from DA to DA right after τ = 0. and the final expression is:   ∂ ∂ ΓL = −2γ x1 + x2 . further- more. From the point of view of DA . once the decomposition of the Kraus map K is given. hence. the Kossakowski-Lindblad vector field generates a one-parameter semigroup of trans- formations describing dissipative quantum dynamics. and V1 = 1 − γ I. 2 the flow of ΓL will not take ρ out of DA for every τ ≥ 0. the vector field YG is automatically defined by the condition G = −V. Consequently. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 39 with G = −V. the explicit form of the dynamics generated by L reads: 1 σ0 + exp(−γτ ) ρ1 σ1 + ρ2 σ2 + ρ3 σ3 . the mathematical tools present in the literature do not always capture relevant physical aspects of some interesting . it is YG = 0. since it is transversal to the spheres centered in x1 = x2 = x3 = 0 representing isospectral states. we can see that. and. Unfortunately. V2 = γ σ3 : L(ρ) = −γ (ρ − σ3 ρ σ3 ) . ZK is the vector field generating the transversal motion which to the strata of DA characterizing dissipative phenomena. Note that. the carrier spaces we are interested in do not possess a “canonical” differential structure. since −G = V = j Vj† Vj = I. 5. (113) This dynamical evolution is known as phase damping.   Φτ (ρ) = (114) 2 and it is clear that this dynamics only affect the phase terms (off-diagonal terms) of ρ represented by the components along σ1 and σ2 . If the initial state 1 2 ρ is a pure state. the flow of ΓL takes a state ρ out of DA . At this purpouse. from the point of view of the space of states DA . for τ ≤ 0. therefore.

G. and G. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 40 Contents situations. Melrose. 7). By means of the vector field ΓL it is possible to describe physical processes. ΓL leads to a semigroup of transformations on DA . [5] J. in which the rank of quantum states is not preserved. Shiva Publishing Limited. Math. Manifolds of differentiable mappings. for the one-dimensional bouncing particle described in section 1. Furthermore. Springer. Cambridge. A. UK. Edin.. 2006. 12:27–29. if the reduction is associated with a quotienting procedure. Cambridge University Press. According with the philosophy exposed in the paper. for a smooth manifold M . we focused on the concrete example of a finite-level quantum system associated with a finite-dimensional C ∗ -algebra A. Berlin. for example. there is an algebra structure at our disposal. Furthermore. In this paper we reviewed some alternative methods to deal with carrier spaces presenting no global differential structure. as it is the case. for instance [5] chap. May 9. Geometry from dynamics. such as decoherence and damping. The advantage of working with the space of functions F(S) rather than with S is that. F(M ) is the algebra of smooth functions. Different mathematical structures defined on S can be directly encoded in the choice of a suitable algebra of function F(S). on F(S). in the algebra F(OA ) of smooth functions on OA . The main tool we adopted is the “duality relation” existing between a set S and the algebra of functions F(S) on it. [3] T. Particular attention has been devoted to the case of manifolds with boundary. References [1] Peter W Michor. then the algebra of functions of the non-trivial carrier spaces becomes a subalgebra of the algebra of functions of the ambient space. and possibly linear.g. The Definition of Lie Derivative. Proc. according to which it is possible to overcome some of the technical difficulties associated to non-trivial carrier spaces by unfolding them into nice. Geometry of Quantum States: An Introduction to Quantum Entanglement. 2015. this physical example suggests that derivations generating groups of transfor- mations are not enough to fully describe physically interesting situations. and thus. Bengtsson and K. Cariñena. the carrier space is the space DA of quantum states which is a stratified manifold with boundary. This procedure enables us ∗ ∗ to define a vector field ΓL on OA (a derivation of F(OA ) in the algebraic language) which is transversal to the strata of DA and is the generator of a semigroup (possi- bly a local one). ambient spaces. which is a linear space. 2003. [2] I. 1960. F. Soc. Ibort.. e. Introduction to microlocal analysis. Morandi. classical and quantum. In this case. the use of F(S) perfectly fits into the so-called unfolding- reduction scheme (see. Marmo. [4] R. J. 1980. Willmore. Lectures at MIT. Zyczkowski. Orpington. For example. we represent the algebra F(DA ) of functions on the space of quantum ∗ ∗ states. . A ∗ suitable ambient space for DA is the dual space OA of the space OA of observables of the theory. Specifically. and we need to consider generators of semigroups of transformations.

Kuś. Marsden. Physique Théorique. Open Systems & Information Dynamics. P. C. Theory of vector-valued differential forms. General theory of Lie groupoids and Lie algebroids. and C. Annales de l’I. Kähler. 13(04):343–362. Homologie et cohomologie des algebres de lie. Lopez-Monsalvo. volume 6. Marmo. 10(08):1360001.Proceedings Supplements. part ii. Graf. Nijenhuis. 1996.. pages 1–108. Rubano. Rhode Island. Landsman. 1984. Grabowski. 2007. Amsterdam. Koszul. Completely positive dynamical . Kus. Grabowski. May 9. N. and G. 1950. [8] G. Nijenhuis. 2005.DG]. G. Marmo. 14(4):355–370. and J. Nettel. Mathematical and conceptual foundations of 20th-century physics. and applica- tions. Facchi. pages 417–553. tensor analysis. Princeton. 2005. Brackets. Ven- triglia. 1995. C ∗ -algebras by example. Grabowski. Elsevier. E. derivations in the graded ring of differential forms. Konikl. 2000. arXiv. [14] J. Gorini. Morandi. Desbrun. 374(48):4801–4803. Sudarshan. On the relation between states and maps in infinite dimensions.. Princeton University Press. [15] G. Addison-Wesley. [20] H. 2007. Akad. 2010. 29(01):85–109. 38(47). almost-complex structures. G. Princeton. 2005. [26] J. New York. Physics Letters A.-L. Phys. [17] A. Marmo and G. 2013. Amsterdam. Ratiu. Kulkarni. Nelson. Elements of Algebraic Topology. The Netherlands. J. Singapore. Provi- dence. 78:65–127. R. Frölicher and A. Marmo. Frölicher and A. Abraham. Landi and G. Munkres. [25] P. group actions. North-Holland. Davidson. [12] A. Proc. World Scientific. Bulletin de la Société Mathématique de France. Morandi. Contact symmetries and hamilto- nian thermodynamics. 1990. [11] A. G. 1962.P. Emch. E. [21] K. and G. Ned. Open Systems and Information Dynamics. Low-Dimensional Quantum Field Theories for Condensed Matter Physicists. Cali- fornia. Der innere differentialkalkül. Manifolds. Man’ko. Springer-Verlag. Marmo. A61:414–429. Mackenzie. [9] E. Proc. 1958. and entanglement. Hirani. Ned. and F. M.Rept. A. Annals of Physics. 1988. Marmo. V. Akad. and T. Konikl. Leok. Bravetti. 21:425–523. R. A: Math. Michelson. C. 1990. Marsden. 2006. B. Tensor analysis. and F. Wetenschap. Menlo Park. Differential forms as spinors. Geometry of quantum systems: density states and entanglement. L. Weten- schap. Princeton University Press. part i.Phys. C.. [29] V. New Jersey. 361:377–400. Nuclear Physics B . [19] W. Lo Vecchio. Rendiconti di Matematica.. Classical and quantum fisher information in the geometrical formulation of quantum mechanics. Theory of vector-valued differential forms. The inverse prob- lem in the calculus of variations and the geometry of the tangent bundle. M. Handbook of the Philosophy of Science. G. [10] J. 18(1):171 – 206. New Jersey. [28] K. Marmo. C. [24] J. and E. Ferrario. A. Gen. 2015. M. Kossakowski. Algebraic differential calculus for gauge theories. Cambridge University Press. R. Sudarshan. chapter Some Geometry and Topology. C. 1956. [22] R. Symmetries. [13] G. Grabowski. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries Contents 41 [6] J. A59:338–359. chapter Between classical and quantum. Cambridge. 1978. 2015. Discrete exterior calculus. [16] N. Spin Geometry. M. and G. J. I. Lawson and M. 188:147–284. Kuś. [27] J. E.H.S. [18] E. math/0508341 [math. 1989. G. volume 100. [23] M. American Mathematical Society. [7] G. International Journal of Geometric Methods in Modern Physics.

American Mathe- matical Society. May 9. On the Generators of Quantum Dynamical Semigroups. [30] G. . 1976. 48:119 – 130. 2017 0:25 WSPC/INSTRUCTION FILE manifolds-with-boundaries 42 Contents semigroups of N-level systems. Journal of Mathematical Physics. Providence. Phillips. 1976. Rhode Island. 1957. S. Communica- tions in Mathematical Physics. [31] E. 17(5):821 – 825. Lindblad. Functional analysis and semigroups. Hille and R.