1. Introduction
Using local notations, this paper is mainly concerned with the following two
connected problems: Given a differential operator
, how can we find
compatibility conditions (CC), that is how can we construct a sequence
such that
and, among all such possible sequences, what
are the “best” ones, at least among the generating ones and when could we say that the sequence obtained is “exact” in a purely formal way, that is using only computer algebra for testing such a property? The order of an operator will be indicated under its arrow.
The difficulty is that, physicists being more familiar with analysis, will say that a sequence is “locally exact” if one can find locally
such that
whenever
. However, they have in mind the property of the exterior derivative d and Maxwell equations in electromagnetism (EM), that is to say, using standard notations, the (local) possibility to introduce the EM potential A such that
whenever the EM field F is a closed 2-form with
.
The main purpose of this paper is to prove that the “things” may be much more delicate and that these problems are only rarely associated with exterior calculus. We use the notations that can be found at length in our many books ( [1] - [6] ) or papers ( [7] [8] [9] [10] [11] ).
Let
be a multi-index with length
, class i if
and
. We set
with
when
. If E is a vector bundle over X with local coordinates
for
and
, we denote by
the q-jet bundle of E with local coordinates simply denoted by
and sections
transforming like the section
when
is an arbitrary section of E. Then both
and
are over
and the Spencer operator, which is defined on sections, just allows to distinguish them by introducing a kind of “difference” through the operator
with local components
and more generally
. In a symbolic way, when changes of coordinates are not involved, it is sometimes useful to write down the components of d in the form
. The restriction of d to the kernel
of the canonical projection
is minus the Spencer map
and
. The kernel of d is made by sections such that
. Finally, if
is a system of order q on E locally defined by linear equations
, the r-prolongation
is locally defined when
by the set of linear equations
,
and has symbol
if one looks at the top order terms. If
is over
, differentiating the identity
with respect to
and substracting the identity
, we obtain the identity
and thus the restriction
( [1] [3] [4] [12] ).
DEFINITION 1.1:
is said to be s-acyclic if the purely algebraic
-cohomology
of
are such that
and involutive if it is n-acyclic. Also
is said to be involutive if it is formally integrable (FI), that is when the restriction
is an epimorphism
or, equivalently, when all the equations of order
are obtained by r prolongations only,
and
is involutive. In that case,
is a canonical equivalent formally integrable first order involutive system on
with no zero order equations, called the Spencer form.
EXAMPLE 1.2: (Classical Killing operator)
Considering the classical Killing operator
where
is the Lie derivative with respect to
and
is a nondegenerate metric with
. Accordingly, it is a lie operator with
and we denote simply by
the set of solutions with
. Now, as we have explained many times, the main problem is to describe the CC of
in the form
by introducing the so-called Riemann operator
. We advise the reader to follow closely the next lines and to imagine why it will not be possible to repeat them for studying the conformal Killing operator. Introducing the well known Levi-Civita isomorphism
by defining the Christoffel
symbols
where
is the inverse matrix of
and the formal Lie derivative of gometric objects, we get the second order system
:
with sections
transforming like
. The system
has a symbol
depending only on
with
and is finite type because its first prolongation is
. It cannot be thus involutive and we need to use one additional prolongation. Indeed, using one of the main results to be found in ( [4] [5] ), we know that, when
is FI, then the CC of
are of order
where s is the number of prolongations needed in order to get a 2-acyclic symbol, that is
in the present situation, a result that should lead to CC of order 2 if
were FI. However, it is known that
is FI, thus involutive, if and only if
has constant Riemannian curvature, a result first found by L.P. Eisenhart in 1926 ( [13] ) which is only a particular example of the Vessiot structure equations discovered by E. Vessiot in 1904 ( [14] ), though in a quite different setting (See [4] and [15] for an explicit modern proof). Such a necessary condition for constructing an exact differential sequence could not have been used by any follower because the “Spencer machinery” has only been known after 1970 ( [12] ). Otherwise, if the metric does not satisfy this condition, CC may exist but have no link with the Riemann tensor ( [10] ). We may define the vector bundle
in the short exact sequence made by the top row of the following commutative diagram:
where the vertical
-sequences are exact but the first, or, using a snake type diagonal chase, from the short exact sequence of vector bundles:
This result is first leading to the long exact sequence of vector bundles:
and to the Riemann operator
. As
, we also discover that
is just the Spencer
-cohomology
at
along the previous short exact sequence.
We get the striking formulas where the + signs are replaced by − signs:
This result, first found as early as in 1978 ( [9] ), clearly exhibit without indices the two well known algebraic properties of the Riemann tensor as a section of the tensor bundle
.
It thus remains to exhibit the Bianchi operator exactly as we did for the Riemann operator, with the same historical comments already provided. However, now we know that
is formally integrable (otherwise nothing could be achieved and we should start with a smaller system [1] [4] [6] ), the construction of the linearized Janet-type differential sequence as a strictly exact differential sequence but not an involutive differential sequence because the system
and thus the first order operator
are formally integrable though not involutive as
is finite type with
but not involutive. Doing one more prolongation
only, we obtain the first order Bianchi operator
as
before, defining the vector bundle
in the long exact sequence made by the top row of the following commutative diagram:
where the vertical
-sequences are exact but the first, or, using a snake type diagonal chase, from the short exact sequence:
showing that
( [8] [9] ). We have in particular for
:
and thus
when
. This result also exhibits all the properties of the Bianchi identities as a section of the tensor bundle
. In arbitrary dimension, we finally obtain the differential sequence, which is not a Janet sequence:
EXAMPLE 1.3: (Conformal Killing operator)
At first sight, it seems that similar methods could work in order to study the conformal Killing operator and, more generally, all conformal concepts will be described with a “hat”, in order to provide the strictly exact differential sequence:
where
is the Weyl operator with generating CC
. It is only in 2016 (see [9] and [15] for more details) that we have been able to recover all these operators and confirm with computer algebra that the orders of the operators involved highly depend on the dimension as follows:
·
:
·
:
·
:
These results are bringing the need to revisit entirely the mathematical foundations of conformal geometry, in particular when
because the Weyl type operator is of third order and when
because the Bianchi type operator is second order in this case contrary to the situation met when
. However, surprisingly, these results have never been acknowledged and the reader will not discover a single reference on such questions in the mathematical literature.
The reason is probably because these results are based on the following technical lemma that could not be even imagined without a deep knowledge and practice of the Spencer
-cohomology (see [16] for details):
LEMMA 1.4: The symbol
defined by the linear equations:
does not depend on any conformal factor, is finite type with
and is surprisingly such that
is 2-acyclic for
or even 3-acyclic when
.
REMARK 1.5: In order to emphasize the reason for using Lie equations, we now provide the explicit form of the n infinitesimal relations with
, whenever
:
where the underlying metric is used for the scalar product
involved. It is easy to check that
defined by
belongs to
with
in the following formula where
is the standard Kronecker symbol and
:
We thus understand how important it is to use “sections” rather than “solutions”.
Accordingly, a possible unification can be achieved through the “fundamental diagram I” relating together the Spencer sequence and the Janet sequence as follows in arbitrary dimension n for any involutive system
because these are the only existing canonical sequences ( [1] ):
where
and
. Indeed, we have
for finite type involutive systems and we therefore notice that the crucial point is to deal with involutive systems. In the group framework, we have
and, as we are dealing with finite type systems, it is thus sufficient to replace
and
by
and
with
in the classical situation or by
and
with
in the conformal situation, on the condition to be able to treat the specific cases
and
.
Finally, as a different way to look at these questions, if K be a differential field containing
, we may introduce the ring
of differential operators with coefficients in K and consider a linear differential operator
with coefficients in K. If
generates the CC of
, we have of course
. Taking the respective (formal) adjoint operators, we obtain therefore
but
may not generate the CC of
and so on in any differential sequence where each operator generates the CC of the preceding one.
DEFINITION 1.6: If M is the differential module over D or simply D-module defined by
, we set
. As for the other extension modules, they have been created in order to “measure” the previous gaps ( [5] ). In particular, we say that
if
generates the CC of
, that
if
generates the CC of
and so on. Moreover, if
is of finite type, then
is surjective with
. The simplest example is that of classical space geometry with
and
. Similar definitions are also valid for the Janet and Spencer sequences. Also, vanishing of the first extension module amounts to the existence of a local parametrization by potential-like functions ( [7] ).
According to a (difficult) theorem of (differential) homological algebra, the extension modules only depend on M and not on the previous differential sequences used ( [17] [18]. They are used in agebraic geometry and have even been introduced in engineering sciences after 1990 (control theory) ( [5] [6] ). However, though the extension modules are the only intrinsic objects that can be associated with a differential module, they have surprisingly never been introduced in mathematical physics. The main problem is that a control system is controllable if and only if it is parametrizable by potentials while the systems involved can be parametrized in all classical physics (Cauchy or Maxwell equations are well known examples in [7] ) apart from... Einstein equations ( [8] ). As for the tools involved, we let the reader compare ( [2] [3] ) to ( [19] [20] ).
After presenting two motivating examples in Section 2, such a procedure will be achieved in Section 3 in such a way that the Spencer sequences involved, being isomorphic to tensor products of the Poincaré sequence for the exterior derivative by finite dimensional Lie algebras, will have therefore vanishing zero, first and second extension modules when
( [4] [11] ). For all results concerning differential modules, we refer the reader to the (difficult) references ( [5] [21] [22] [23] ).
2. Two Motivating Examples
EXAMPLE 2.1
With
, let us consider the inhomogeneous second order operator:
We obtain at once through crossed derivatives:
and, by substituting, two fourth order CC for
, namely:
However, the commutation relation
provides a single CC for
, namely:
and we check at once
while
, hat is:
Using corresponding notations, let us compare the two following differential sequences:
Though the second order system considered is surely not FI because the 4 parametric jets of
are
and the 4 (again !) parametric jets of
are
but the 4 (again !) parametric jets of
are
. More generally, we let the reader prove by induction that
. The formal r-prolongation of (2), namely:
is exact because
, even though the corresponding symbol sequence:
is not exact because
because the system considered is not formally integrable.
On the contrary, the prolongations of (1) are not exact on the jet level. Indeed, the long sequence:
is not exact because we have
.
Now, considering the ring
of differential operators with coefficients in the trivial differential field
, we have the “exact” sequences of differential modules where
:
where p is the canonical residual projection. However, and this is a quite delicate point rarely known even by mathematicians, a fortiori by physicists, they are not “strictly” exact even if the Euler-Poincaré characteristics both vanish because
and
(see [15] for definitions and more details). Roughly speaking, it follows that the “best” differential sequences are obtained by using only formally integrable operators/systems in such a way that sequences on the jet level can be studied through their symbol sequences, the “canonical” ones by using exclusively involutive operators/systems in such a way that what happens with
also happens with
and so on. It follows that the sequences (2) or (2*) are “better” than (1) or (1*) because they provide more information on the generating CC.
However, the given system is not FI and it should be “better” to use another system providing more information. In particular, if we start wth a system
and set
, it is known that (in general) one can find two integers
such that the system
is formally integrable and even involutive with the same solutions ( [1] [5] [6] ). When all the operators are FI, the sequence is said to be strictly exact ( [24] ).
In the present situation, it should be “better” to replace
by
because
is adding
while
is adding
and
is adding
. It follows that the Janet sequence for the injective trivially involutive operator
is providing even more information, along with the fact that the Spencer bundles vanish in the “fundamental diagram I” ( [1] [4] [5] ).
We let the reader check that all the extension modules vanish because
and to compare with the totally different involutive system defined by
with
,
,
.
EXAMPLE 2.2
· FIRST STEP With
, let us consider the second order linear system
introduced by F.S. Macaulay in his 1916 book ( [25] ) (See also [6] for more details):
Using muli-indices, we may introduce the operators
. Taking into account the 3 commutation relations
and the single Jacobi identity
, we obtain at once the following locally and strictly exact sequence where the order of each operator is under its own arrow:
However, the first operator
involved cannot be involutive because it is finite type, that is
for a certain integer
as we must have an exact sequence
and so on. The first prolongation is obtained by adding the 9 PD equations:
and the second prolongation is obtained by adding the 15 PD equations
. We obtain therefore
,
,
. Nevertheless, the interesting fact is that
is 2-acyclic without being 3-acycic and thus involutive. Indeed, we have the
-sequences:
Using the letter v for the symbol coordinates, the mapping
on the left is defined by:
that is to say
,
,
. The corresponding
-map is thus injective and surjective, that is
is 2-acyclic but cannot be also 3-acyclic because of the inequality,
. The above sequence is thus very far from being a Janet sequence and we cannot compare it with the Spencer sequence.
· SECOND STEP In the example of Macaulay, we have at once
with the 7 parametric jets
and thus
with the only additional third order parametric jet (
). We notice that, when
, the new system
defined by
,
is also finite type with
and thus
and we invite the reader to treat directly such an elementary example as an exercise and to compare (see [25] for this striking result on the powers of 2). Therefore, instead of starting with the previous second order operator
defined by
, we may now start afresh with the new third order operator
defined by
which is not involutive again. We let the reader check as a tricky exercise or using computer algebra that one may obtain “necessarily” the following finite length differential sequence which is far from being a Janet sequence but for other reasons.
and we check that
. As
is 2-acyclic, the third order operator
has a CC operator
of order 1 having a CC operator
of order 2 which is involutive, totally by chance, and we end with the Janet sequence for
. Such a situation is the only one we have met during the last... 40 years !. (see [15], p 119-126 for more details).
· THIRD STEP We may finally start with the new operator
defined by the involutive system
with symbol
. The following “fundamental diagram I” only depends on its left commutative square and
. Each horizontal sequence is formally exact and can be constructed step by step. The interest is that we have
because
. It is nevertheless, even today, not so well known that the three differential sequences appearing in this diagram can be constructed “step by step” or “as a whole” ( [1] [4] [5] [6] ). Accordingly, the reader not familiar with the formal theory of systems of PD equations may find difficult to deal with the following definitions of the Spencer bundles
and Janet bundles
for an involutive system
of order q over E:
For this reason, we prefer to use successive compatibility conditions, starting from the commutative square
on the left of the next diagram. The Janet tabular of the Macaulay system and its prolongations up to order 4 can be decomposed as follows ( [26] ):
The total number of different single “dots” provides the
CC
.
The total number of different couples of “dots” provides the
CC
.
The total number of different triples of “dots” provides the
CC
.
We obtain therefore the fiber dimensions of the successive Janet bundles in the Janet sequence.
The same procedure can be applied to the Spencer bundles in the Spencer sequence by introducing the new 8 parametric jet indeterminates:
in the first order system defined by 24 PD equations (8 of class 3 + 8 of class 2 + 8 of class 1):
The morphisms
in the vertical short exact sequences are inductively induced from the morphism
in the first short exact vertical sequence on the left. The central horizontal sequence can be called “hybrid sequence” because it is at the same time a Spencer sequence for the first order system
over
and a Janet sequence for the involutive injective operator
. It can be constructed step by step, starting with the short exact sequence:
In actual practice, as the system
is homogeneous, it is thus formally integrable and finite type because the system
is trivially involutive with a symbol
. Accordingly,
is an involutive operator of order 4 and we obtain a finite length Janet sequence which is formally exact both on the jet level and on the symbol level, that can only contain the successive first order operators
. For example, one can determine
just by counting the dimensions, either in the long exact jet sequence:
and obtain
.
However, one can also use the fact that
and
while introducing the restriction
of
to
in the long exact symbol sequence:
in order to obtain again
.
We wish good luck to anybody using Computer Algebra because one should have to deal with a matrix
in order to describe the prolongation morphism
. Nevertheless, in order to give a hint, we recall the vanishing of the Euler-Poincaré characteristic as we can check successively:
In the case of finite type systems, the usefulness of the Spencer sequence is so evident, like on such an example, that it needs no comment.
We invite the reader to treat separately but similarly the system:
and to compare the various extension modules.
3. Solution
According to the previous sections, it only remains to consider the two cases
and
. For simplicity, we shall only consider the situation of the Euclidean metric and the corresponding linear systems. We let the reader treat by himself the nonlinear counterparts.
· CASE
With
, we may consider a section
and introduce the classical Killing system
by means of the formal Lie derivative:
Similarly, with the Christoffel symbol
, we may consider:
The conformal Killing system can be defined with a conformal factor as:
and its first prolongation becomes:
The elimination of
or
does not provide any OD equation of order 1 or 2. Moreover, we let the reader check that
as a way to understand the part plaid by the Spencer operator and the reason for introducing
. With more details, dividing the Killing system by
, we get
. Differentiating this OD equation, we get:
and we just need to substract the OD equation
in order to get:
In order to escape from the previous situation while having a vanishing symbol
, we may consider the new unusual prolongation:
and substract the second order OD equation
multiplied by
while introducing the new geometric object
in order to obtain the third order infinitesimal Lie equation:
The nonlinear framework, not known today because the work of Vessiot is still not acknowledged, explains the successive inclusions
. Indeed, if we consider the translation group
and the bigger isometry group
, the inclusion of groups of the real line:
with the respective finite Lie equations in Lie form with the jet coordinates
:
where we recognize the Schwarzian third order differential invariant of the projective group.
Of course, we have
and the respective linearizations:
The Janet tabular of the conformal system order 3 can be decomposed as follows:
The total number of different single “dots” provides the 0 CC
.
We obtain therefore the fiber dimensions of the successive Janet bundles in the Janet sequence.
The same procedure can be applied to the other canonical differential sequences.
When
, one has 3 parameters (1 translation + 1 dilatation + 1 elation) and we get the following “fundamental diagram I “ only depending on the left commutative square:
In this diagram, the operator
has compatibility conditions
induced by d and the space of solutions
of
is generated over the constants by the three infinitesimal generators:
(translation),
(dilatation),
(elation)
of the action and coincides with the projective group of the real line.
· CASE
The classical approach is to consider the infinitesimal conformal Killing system for
and eliminate the infinitesimal conformal factor
as follows by introducing the formal and the effective Lie derivatives such that
:
that is to say the elimination of A is just producing locally the two well known Cauchy-Riemann equations allowing to define infinitesimal complex transformations of the plane, that is to say an infinite dimensional Lie pseudogroup which is by no way providing a finite dimensional Lie group. As such an operator has no compatibility condition (CC), we obtain by one prolongation
second order equations but another prolongation does not provide a zero symbol at order 3 and it is just such a delicate step that we have to overcome by adding
homogeneous third order PD equations. The only possibility is to consider the following system and to prove that it is defining a system of infinitesimal Lie equations leading to
infinitesimal generators.
where the 4 second order PD equations can also be rewritten with
as:
The general solution of the 8 third order PD equations can be written with 12 arbitrary constant parameters as:
Taking into account the first and second order PD equations, we must have the relations:
and the final number of parameters is indeed reduced to
arbitrary parameters. Collecting the above results, we obtain the 6 infinitesimal generators:
We find back the two infinitesimal generators of the elations, namely:
and
obtained by exchanging
with
.
Contrary to the situation met when
where one starts with a groupoid of order 1 and obtains groupoids of order 2 or 3 after one or two prolongations, in the present situation, we have to check directly the commutation relations for the six infinitesimal generators already found, namely:
We have thus obtained in an unexpected way the desired 2 translations, 1 rotation, 1 dilatation and 2 elations of the conformal group when
.
At order one, we may consider the classical Killing system
obtained by preserving
, the Weyl system
and the conformal system
with
and
. At order two, we have the strict inclusions
with
preserving
,
obtained by preserving
and
obtained by preserving
. The main difference with the case
is that now
has a symbol
,
has also a symbol
but that
with strict inclusion in order to have now
, even though
. However, we are now able to deal with three trivially involutive systems having zero symbols and we have the strict inclusions
with respective dimensions
according to the basic inequalities
valid in arbitrary dimension
. The interest of this result is that we have for the Spencer bundles the strict inclusions
of the zero Spencer bundles, leading to the strict inclusions of the respective linear Spencer sequences because:
in agrement with many recent results ( [21] [22] [23] [24] ). As in Example 2.2, we let the reader introduce the 6 parametric jet indeterminates
.
The Janet tabular of the conformal Killing system and its prolongations up to order 3 can be decomposed as follows:
The total number of different single “dots” provides the
CC
.
The total number of different couples of “dots” provides the
CC
.
We obtain therefore the fiber dimensions of the successive Janet bundles in the Janet sequence.
The same procedure can be applied to the other canonical differential sequences.
When
, one has 6 parameters (2 translations + 1 rotation + 1 dilatation + 2 elations) and we get the following “fundamental diagram I” only depending on the left commutative square:
· CASE
The Janet tabular of the conformal Killing system and its prolongations up to order 3 can be decomposed as follows:
The total number of different single “dots” provides the
CC
.
The total number of different couples of “dots” provides the
CC
.
The total number of different triples of “dots” provides the
CC
.
We obtain therefore the fiber dimensions of the successive Janet bundles in the Janet sequence.
The same procedure can be applied to the other canonical differential sequences and we get the desired “fundamental diagram I” below:
We have 10 parameters (3 translations, 3 rotations, 1 dilataion, 3 elations).
The computation of
needs to determine the rank of a
matrix !
4. Conclusion
We have shown that the true important specific property of the conformal group, at least for applications to physics, is that, even if it is defined as a specific Lie pseudogroup of transformations, it is in fact a Lie group of transformations with a finite number
of parameters or infinitesimal generators when
. Accordingly, in dimension
, we have no OD equation of order 1 and 2, a result leading therefore to add 1 unexpected OD equation of order 3. Similarly, when
, we obtain the Cauchy-Riemann PD equations defining an infinite dimensional Lie pseudogroup and we have therefore to add, again in a totally unexpected way, as many third order PD equations as the number of jet coordinates of strict order 3. When
, the fact that the analogue of the Weyl operator for describing the CC of the conformal operator is of order 3 is rather un-pleasant but this is nothing compared to the fact that, when
, the analogue of the Bianchi operator for describing the CC of the previous second order CC playing the part of the Weyl CC is of order 2 again. And we don’t speak about the case
( [9] [15] ). Though these results can be checked by means of computer algebra and are confirmed by the use of the fundamental diagram I, they do not seem to be known today. Accordingly, any physical theory (existence of gravitational waves or black holes... ) which is not coherent with differential homological algebra (vanishing of the first and second extension modules for the Poincaré sequence in the previous examples...) must be revisited in the light of these new mathematical tools, even if it seems apparently well established ( [8] [27] [28] [29] [30] ).