Existence and Construction of Vessiot Connections
A rigorous formulation of Vessiot's vector field approach to the analysis of general systems of partial differential equations is provided. It is shown that this approach is equivalent to the formal theory of differential equations and that it can be carried through if, and only if, the given s...
Gespeichert in:
| Datum: | 2009 |
|---|---|
| Hauptverfasser: | , |
| Format: | Artikel |
| Sprache: | English |
| Veröffentlicht: |
Інститут математики НАН України
2009
|
| Schriftenreihe: | Symmetry, Integrability and Geometry: Methods and Applications |
| Online Zugang: | http://dspace.nbuv.gov.ua/handle/123456789/149123 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Назва журналу: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Zitieren: | Existence and Construction of Vessiot Connections / D. Fesser, W.M. Seiler // Symmetry, Integrability and Geometry: Methods and Applications. — 2009. — Т. 5. — Бібліогр.: 28 назв. — англ. |
Institution
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
irk-123456789-149123 |
|---|---|
| record_format |
dspace |
| spelling |
irk-123456789-1491232019-02-20T01:26:26Z Existence and Construction of Vessiot Connections Fesser, D. Seiler, W.M. A rigorous formulation of Vessiot's vector field approach to the analysis of general systems of partial differential equations is provided. It is shown that this approach is equivalent to the formal theory of differential equations and that it can be carried through if, and only if, the given system is involutive. As a by-product, we provide a novel characterisation of transversal integral elements via the contact map. 2009 Article Existence and Construction of Vessiot Connections / D. Fesser, W.M. Seiler // Symmetry, Integrability and Geometry: Methods and Applications. — 2009. — Т. 5. — Бібліогр.: 28 назв. — англ. 1815-0659 2000 Mathematics Subject Classification: 35A07; 35A30; 35N99; 58A20 http://dspace.nbuv.gov.ua/handle/123456789/149123 en Symmetry, Integrability and Geometry: Methods and Applications Інститут математики НАН України |
| institution |
Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| collection |
DSpace DC |
| language |
English |
| description |
A rigorous formulation of Vessiot's vector field approach to the analysis of general systems of partial differential equations is provided. It is shown that this approach is equivalent to the formal theory of differential equations and that it can be carried through if, and only if, the given system is involutive. As a by-product, we provide a novel characterisation of transversal integral elements via the contact map. |
| format |
Article |
| author |
Fesser, D. Seiler, W.M. |
| spellingShingle |
Fesser, D. Seiler, W.M. Existence and Construction of Vessiot Connections Symmetry, Integrability and Geometry: Methods and Applications |
| author_facet |
Fesser, D. Seiler, W.M. |
| author_sort |
Fesser, D. |
| title |
Existence and Construction of Vessiot Connections |
| title_short |
Existence and Construction of Vessiot Connections |
| title_full |
Existence and Construction of Vessiot Connections |
| title_fullStr |
Existence and Construction of Vessiot Connections |
| title_full_unstemmed |
Existence and Construction of Vessiot Connections |
| title_sort |
existence and construction of vessiot connections |
| publisher |
Інститут математики НАН України |
| publishDate |
2009 |
| url |
http://dspace.nbuv.gov.ua/handle/123456789/149123 |
| citation_txt |
Existence and Construction of Vessiot Connections / D. Fesser, W.M. Seiler // Symmetry, Integrability and Geometry: Methods and Applications. — 2009. — Т. 5. — Бібліогр.: 28 назв. — англ. |
| series |
Symmetry, Integrability and Geometry: Methods and Applications |
| work_keys_str_mv |
AT fesserd existenceandconstructionofvessiotconnections AT seilerwm existenceandconstructionofvessiotconnections |
| first_indexed |
2025-07-12T21:25:36Z |
| last_indexed |
2025-07-12T21:25:36Z |
| _version_ |
1837477970877349888 |
| fulltext |
Symmetry, Integrability and Geometry: Methods and Applications SIGMA 5 (2009), 092, 41 pages
Existence and Construction of Vessiot Connections?
Dirk FESSER † and Werner M. SEILER ‡
† IWR, Universität Heidelberg, INF 368, 69120 Heidelberg, Germany
E-mail: dirk.fesser@iwr.uni-heidelberg.de
‡ AG “Computational Mathematics”, Universität Kassel, 34132 Kassel, Germany
E-mail: seiler@mathematik.uni-kassel.de
URL: http://www.mathematik.uni-kassel.de/∼seiler/
Received May 05, 2009, in final form September 14, 2009; Published online September 25, 2009
doi:10.3842/SIGMA.2009.092
Abstract. A rigorous formulation of Vessiot’s vector field approach to the analysis of ge-
neral systems of partial differential equations is provided. It is shown that this approach is
equivalent to the formal theory of differential equations and that it can be carried through
if, and only if, the given system is involutive. As a by-product, we provide a novel charac-
terisation of transversal integral elements via the contact map.
Key words: formal integrability; integral element; involution; partial differential equation;
Vessiot connection; Vessiot distribution
2000 Mathematics Subject Classification: 35A07; 35A30; 35N99; 58A20
Contents
1 Introduction 2
2 The contact structure 3
3 The formal theory of differential equations 4
4 The Cartan normal form 7
5 The Vessiot distribution 10
6 Flat Vessiot connections 19
7 The existence theorem for integral distributions 21
8 The existence theorem for flat Vessiot connections 24
9 Conclusions 26
A Proof of Theorem 2 27
B Proof of Proposition 7 35
C Proof of Theorem 3 37
References 40
?This paper is a contribution to the Special Issue “Élie Cartan and Differential Geometry”. The full collection
is available at http://www.emis.de/journals/SIGMA/Cartan.html
mailto:dirk.fesser@iwr.uni-heidelberg.de
mailto:seiler@mathematik.uni-kassel.de
http://www.mathematik.uni-kassel.de/~seiler/
http://dx.doi.org/10.3842/SIGMA.2009.092
http://www.emis.de/journals/SIGMA/Cartan.html
2 D. Fesser and W.M. Seiler
1 Introduction
Constructing solutions for (systems of) partial differential equations is obviously difficult – in
particular for non-linear systems. Élie Cartan [4] proposed to construct first infinitesimal solu-
tions or integral elements. These are possible tangent spaces to (prolonged) solutions. Thus they
always lead to a linearisation of the problem and their explicit construction requires essentially
only straightforward linear algebra. In the Cartan–Kähler theory [3, 12], differential equations
are represented by exterior differential systems and integral elements consist of tangent vectors
pointwise annihilated by differential forms.
Vessiot [28] proposed in the 1920s a dual approach which does not require the use of exterior
differential systems. Instead of individual integral elements, it always considers distributions
of them generated by vector fields and their Lie brackets replace the exterior derivatives of
differential forms. This approach takes an intermediate position between the formal theory of
differential equations [13, 19, 23] and the Cartan–Kähler theory of exterior differential systems.
Thus it allows for the transfer of many techniques from the latter to the former one, although
this point will not be studied here.
Vessiot’s approach may be considered a generalisation of the Frobenius theorem. Indeed, if
one applies his theory to a differential equation of finite type, then one obtains an involutive
distribution such that its integral manifolds are in a one-to-one correspondence with the smooth
solutions of the equation. For more general equations, Vessiot proposed to “cover” the equation
with infinitely many involutive distributions such that any smooth solution corresponds to an
integral manifold of at least one of them.
Vessiot’s theory has not attracted much attention: presentations in a more modern language
are contained in [5, 24]; applications have mainly appeared in the context of the Darboux
method for solving hyperbolic equations, see for example [27]. While a number of textbooks
provide a very rigorous analysis of the Cartan–Kähler theory, the above mentioned references
(including Vessiot’s original work [28]) are somewhat lacking in this respect. In particular, the
question of under what assumptions Vessiot’s construction succeeds has been ignored.
The purpose of the present article is to close this gap and at the same time to relate the Vessiot
theory with the key concepts of the formal theory like formal integrability and involution (we
will not develop it as a dual form of the Cartan–Kähler theory, but from scratch within the
formal theory). We will show that Vessiot’s construction succeeds if, and only if, it is applied to
an involutive system of differential equations. This result is of course not surprising, given the
well-known fact that the formal theory and the Cartan–Kähler theory are equivalent. However,
to our knowledge an explicit proof has never been given. As a by-product, we will provide
a new characterisation of integral elements based on the contact map, making also the relations
between the formal theory and the Cartan–Kähler theory more transparent. Furthermore, we
simplify the construction of the integral distributions. Up to now, quadratic equations had to
be considered (under some assumptions, their solution can be obtained via a sequence of linear
systems. We will show how the natural geometry of the jet bundle hierarchy can be exploited
for always obtaining a linear system of equations.
This article contains the main results of the first author’s doctoral thesis [6] (a short summary
of the results has already appeared in [7]). It is organised as follows. The next three sections
recall the needed elements from the formal theory of differential equations and provide our new
characterisation of transversal integral elements. The following two sections introduce the key
concepts of Vessiot’s approach: the Vessiot distribution, integral distributions and flat Vessiot
connections. Two further sections discuss existence theorems for the latter two based on a step-
by-step approach already proposed by Vessiot. As the proofs of some results are fairly technical,
the main text contains only an outline of the underlying ideas and full details are given in three
appendices. The Einstein convention is sometimes used to indicate summation.
Existence and Construction of Vessiot Connections 3
2 The contact structure
Before we outline the formal theory of partial differential equations, we briefly review its under-
lying geometry: the jet bundle and its contact structure. Many different ways exist to introduce
these geometric constructions, see for example [9, 18, 20]. Furthermore, they are discussed in
any book on the formal theory (see the references in the next section).
Let π : E → X be a smooth fibred manifold. We call coordinates x = (xi : 1 ≤ i ≤ n) of X
independent variables and fibre coordinates u = (uα : 1 ≤ α ≤ m) in E dependent variables.
Sections σ : X → E correspond locally to functions u = s(x). We will use throughout a “global”
notation in order to avoid the introduction of many local neighbourhoods even though we mostly
consider local sections.
Derivatives are written in the form uα
µ = ∂|µ|uα/∂xµ1
1 · · · ∂xµn
n where µ = (µ1, . . . , µn) is
a multi-index. The set of derivatives uα
µ up to order q is denoted by u(q); it defines a local
coordinate system for the q-th order jet bundle Jqπ, which may be regarded as the space of
truncated Taylor expansions of functions s.
The hierarchy of jet bundles Jqπ with q = 0, 1, 2, . . . possesses many natural fibrations
which correspond to “forgetting” higher-order derivatives. For us particularly important are
πq
q−1 : Jqπ → Jq−1π and πq : Jqπ → X . To each section σ : X → E , locally defined by
σ(x) =
(
x, s(x)
)
, we may associate its prolongation jqσ : X → Jqπ, a section of the fibration πq
locally given by jqσ(x) =
(
x, s(x), ∂xs(x), ∂xxs(x), . . .
)
.
The geometry of the jet bundle Jqπ is to a large extent determined by its contact structure. It
can be introduced in various ways. For our purposes, three different approaches are convenient.
First, we adopt the contact codistribution C0
q ⊆ T ∗(Jqπ); it consists of all one-forms such that
their pull-back by a prolonged section vanishes. Locally, it is spanned by the contact forms
ωα
µ = duα
µ −
n∑
i=1
uα
µ+1i
dxi, 0 ≤ |µ| < q, 1 ≤ α ≤ m.
Dually, we may consider the contact distribution Cq ⊆ T (Jqπ) consisting of all vector fields
annihilated by C0
q . A straightforward calculation shows that it is generated by the contact fields
C
(q)
i = ∂i +
m∑
α=1
∑
0≤|µ|<q
uα
µ+1i
∂uα
µ
, 1 ≤ i ≤ n,
Cµ
α = ∂uα
µ
, |µ| = q, 1 ≤ α ≤ m. (1)
Note that the latter fields, Cµ
α , span the vertical bundle V πq
q−1 of the fibration πq
q−1. Thus
the contact distribution can be split into Cq = V πq
q−1 ⊕ H. Here the complement H is an n-
dimensional transversal subbundle of T (Jqπ) and obviously not uniquely determined (though
any local coordinate chart induces via the span of the vectors C
(q)
i one possible choice). Any
such complement H may be considered the horizontal bundle of a connection on the fibred
manifold πq : Jqπ → X (but not for the fibration πq
q−1). Following Fackerell [5], we call any
connection on πq the horizontal bundle of which consists of contact fields a Vessiot connection
(in the literature the terminology Cartan connection is also common, see for example [15]).
For later use, we note the structure equations of the contact distribution. The only non-
vanishing Lie brackets of the vector fields (1) are[
Cν+1i
α , C
(q)
i
]
= ∂uα
ν
, |ν| = q − 1. (2)
Note that this observation implies that the vertical bundle V πq
q−1 is involutive.
4 D. Fesser and W.M. Seiler
As a third approach to the contact structure we consider, following Modugno [17], the contact
map (see also [6, 23]). It is the unique map Γq : Jqπ ×
X
TX → T (Jq−1π) such that the diagram
Jqπ ×
X
TX Γq // T (Jq−1π)
TX
((jqσ)◦τX )×idTX
ddIIIIIIIII T (jq−1σ)
::tttttttttt
commutes for any section σ. Because of its linearity over πq
q−1, we may also consider it a map
Γq : Jqπ → T ∗X ⊗
Jq−1π
T (Jq−1π) with the local coordinate form
Γq : (x,u(q)) 7→
(
x,u(q−1); dxi ⊗ (∂xi + uα
µ+1i
∂uα
µ
)
)
. (3)
Obviously, Γq(ρ, ∂xi) = Tρπ
q
q−1(C
(q)
i ) and hence (Cq)ρ = (Tρπ
q
q−1)
−1
(
im Γq(ρ)
)
for any point
ρ ∈ Jqπ. Note that all vectors in the image of Γq(ρ) are transversal to the fibration πq−1
q−2.
One of the main applications of the contact structure is given by the following proposition
(for a proof, see [6, Proposition 2.1.6] or [23, Proposition 2.2.7]). It characterises those sections
of the jet bundle πq : Jqπ → X which are prolongations of sections of the underlying fibred
manifold π : E → X .
Proposition 1. A section γ : X → Jqπ is of the form γ = jqσ for a section σ : X → E if, and
only if, im Γq
(
γ(x)
)
= Tγ(x)π
q
q−1
(
Tγ(x) im γ
)
for all points x ∈ X where γ is defined.
Thus for any section σ : X → E the equality im Γq+1
(
jq+1σ(x)
)
= im Tx(jqσ) holds and we
may say that knowing the (q + 1)-jet jq+1σ(x) of a section σ at some x ∈ X is equivalent to
knowing its q-jet ρ = jqσ(x) at x as well as the tangent space Tρ(im jqσ) at this point. This
observation will later be the key for the Vessiot theory.
3 The formal theory of differential equations
We are now going to outline the formal theory of partial differential equations to introduce the
basic notation. Our presentation follows [23]; other general references are [13, 14, 19].
Definition 1. A differential equation of order q is a fibred submanifold Rq ⊆ Jqπ locally
described as the zero set of some smooth functions on Jqπ:
Rq :
{
Φτ
(
x,u(q)
)
= 0,
(τ = 1, . . . , t).
(4)
Note that we do not distinguish between scalar equations and systems.
We denote by ι : Rq ↪→ Jqπ the canonical inclusion map. Differentiating every equation
in the local representation (4) leads to the prolonged equation Rq+1 ⊆ Jq+1π defined by the
equations Φτ = 0 and DiΦ
τ = 0 where the formal derivative Di is given by
DiΦ
τ
(
x,u(q+1)
)
=
∂Φτ
∂xi
(
x,u(q)
)
+
∑
0≤|µ|≤q
m∑
α=1
∂Φτ
∂uα
µ
(
x,u(q)
)
uα
µ+1i
. (5)
Iteration of this process gives the higher prolongations Rq+r ⊆ Jq+rπ. A subsequent projec-
tion leads to R(1)
q = πq+1
q (Rq+1) ⊆ Rq, which is a proper submanifold whenever integrability
conditions appear.
Existence and Construction of Vessiot Connections 5
Definition 2. A differential equation Rq is formally integrable if at any prolongation order
r > 0 the equality R(1)
q+r = Rq+r holds.
In local coordinates, the following definition coincides with the usual notion of a solution.
Definition 3. A solution is a section σ : X → E such that its prolongation satisfies im jqσ ⊆ Rq.
For formally integrable equations it is straightforward to construct order by order formal
power series solutions. Otherwise it is hard to find solutions. A constitutive insight of Cartan
was to introduce infinitesimal solutions or integral elements at a point ρ ∈ Rq as subspaces
Uρ ⊆ TρRq which are potentially part of the tangent space of a prolonged solution.
Definition 4. Let Rq ⊆ Jqπ be a differential equation and ι : Rq → Jqπ the canonical inclusion
map. Let I[Rq] = 〈ι∗C0
q 〉diff be the differential ideal generated by the pull-back of the contact
codistribution on Rq (algebraically, I[Rq] is then spanned by a basis of ι∗C0
q and the exterior
derivatives of the forms in this basis). A linear subspace Uρ ⊆ TρRq is an integral element at
the point ρ ∈ Rq, if all forms in (I[Rq])ρ vanish on it.
The following result provides an alternative characterisation of transversal integral elements
via the contact map. It requires that the projection πq+1
q : Rq+1 → Rq is surjective.
Proposition 2. Let Rq be a differential equation such that R(1)
q = Rq. A linear subspace
Uρ ⊆ TρRq such that Tρι(Uρ) lies transversal to the fibration πq
q−1 is an integral element at the
point ρ ∈ Rq if, and only if, a point ρ̂ ∈ Rq+1 exists on the prolonged equation Rq+1 such that
πq+1
q (ρ̂) = ρ and Tρι(Uρ) ⊆ im Γq+1(ρ̂).
Proof. Assume first that Uρ satisfies the given conditions. It follows immediately from the
coordinate form of the contact map that then firstly Tρι(Uρ) is transversal to πq
q−1 and secondly
that every one-form ω ∈ ι∗C0
q vanishes on Uρ, as im Γq+1(ρ̂) ⊂ (Cq)ρ. Thus there only remains
to show that the same is true for the two-forms dω ∈ ι∗(dC0
q ).
Choose a section γ : Rq → Rq+1 such that γ(ρ) = ρ̂ and define a distribution D of rank
n on Rq by setting Tι(Dρ̃) = im Γq+1
(
γ(ρ̃)
)
for any point ρ̃ ∈ Rq. Obviously, by construction
Uρ ⊆ Dρ. It follows now from the coordinate form (3) of the contact map that locally the
distribution D is spanned by n vector fields Xi such that ι∗Xi = C
(q)
i + γα
µ+1i
Cµ
α where the
coefficients γα
ν are the highest-order components of the section γ. Thus the commutator of two
such vector fields satisfies
ι∗
(
[Xi, Xj ]
)
=
(
C
(q)
i (γα
µ+1j
)− C
(q)
j (γα
µ+1i
)
)
Cµ
α + γα
µ+1j
[C(q)
i , Cµ
α ]− γα
µ+1i
[C(q)
j , Cµ
α ].
The commutators on the right hand side vanish whenever µi = 0 or µj = 0, respectively.
Otherwise we obtain −∂uα
µ−1i
and −∂uα
µ−1j
, respectively. But this fact implies that the two sums
on the right hand side cancel each other and we find that ι∗
(
[Xi, Xj ]
)
∈ Cq. Thus we find for
any contact form ω ∈ C0
q that
ι∗(dω)(Xi, Xj) = dω(ι∗Xi, ι∗Xj) = ι∗Xi
(
ω(ι∗Xj)
)
− ι∗Xj
(
ω(ι∗Xi)
)
+ ω
(
ι∗([Xi, Xj ])
)
.
Each summand in the last expression vanishes, as all appearing fields are contact fields. Hence
any form ω ∈ ι∗(dC0
q ) vanishes on D and in particular on Uρ ⊆ Dρ.
For the converse, note that any transversal integral element Uρ ⊆ TρRq is spanned by linear
combinations of vectors vi such that Tρι(vi) = C
(q)
i |ρ + γα
µ,iC
µ
α |ρ where γα
µ,i are real coefficients.
Now consider a contact form ωα
ν with |ν| = q − 1. Then dωα
ν = dxi ∧ duα
ν+1i
. Evaluating
the condition ι∗(dωα
ν )|ρ(vi, vj) = dω
(
Tρι(vi), Tρι(vj)
)
= 0 yields the equation γα
ν+1i,j
= γα
ν+1j ,i.
Hence the coefficients are of the form γα
µ,i = γα
µ+1i
and a section σ exists such that ρ = jqσ(x)
and Tρ(im jqσ) is spanned by the vectors Tρι(v1), . . . , Tρι(vn). This observation implies that Uρ
satisfies the given conditions. �
6 D. Fesser and W.M. Seiler
For many purposes the purely geometric notion of formal integrability is not sufficient, and
one needs the stronger algebraic concept of involution. This concerns in particular the derivation
of uniqueness results but also the numerical integration of overdetermined systems [21]. An
intrinsic definition of involution is possible using the Spencer cohomology (see for example [22]
and references therein for a discussion). We apply here a simpler approach requiring that one
works in “good”, more precisely: δ-regular, coordinates x. This assumption represents a mild
restriction, as generic coordinates are δ-regular and it is possible to construct systematically
“good” coordinates – see [11]. Furthermore, it will turn out that the use of δ-regular coordinates
is essential for Vessiot’s approach.
Definition 5. The (geometric) symbol of a differential equation Rq is Nq = V πq
q−1|Rq ∩ TRq.
Thus, the symbol is the vertical part of the tangent space to Rq. Locally, Nq consists of all
vertical vector fields
∑m
α=1
∑
|µ|=q vα
µ∂uα
µ
where the coefficients vα
µ satisfy the following linear
system of algebraic equations:
Nq :
m∑
α=1
∑
|µ|=q
(
∂Φτ
∂uα
µ
)
vα
µ = 0. (6)
The matrix of this system is called the symbol matrix Mq. The prolonged symbols Nq+r are the
symbols of the prolonged equations Rq+r with corresponding symbol matrices Mq+r.
The class of a multi-index µ = (µ1, . . . , µn), denoted cls µ, is the smallest k for which µk is
different from zero. The columns of the symbol matrix (6) are labelled by the vα
µ . We order them
as follows. Let α and β denote indices for the dependent coordinates, and let µ and ν denote
multi-indices for marking derivatives. Derivatives of higher order are greater than derivatives
of lower order: if |µ| < |ν|, then uα
µ ≺ uβ
ν . If derivatives have the same order |µ| = |ν|, then we
distinguish two cases: if the leftmost non-vanishing entry in µ−ν is positive, then uα
µ ≺ uβ
ν ; and
if µ = ν and α < β, then uα
µ ≺ uβ
ν . This is a class-respecting order: if |µ| = |ν| and cls µ < cls ν,
then uα
µ ≺ uβ
ν . Any set of objects indexed with pairs (α, µ) can be ordered in an analogous way.
This order of the multi-indices µ and ν is called the degree reverse lexicographic ranking, and
we generalise it in such a way that it places more weight on the multi-indices µ and ν than on
the numbers α and β of the dependent variables. This is called the term-over-position lift of the
degree reverse lexicographic ranking.
Now the columns within the symbol matrix are ordered descendingly according to the degree
reverse lexicographic ranking for the multi-indices µ of the variables vα
µ in equation (6) and
labelled by the pairs (α, µ). (It follows that, if vα
µ and vβ
ν are such that cls µ > cls ν, then the
column corresponding to vα
µ is left of the column corresponding to vβ
ν .) The rows are ordered in
the same way with regard to the pairs (α, µ) of the variables uα
µ which define the classes of the
equations Φτ (x,u(q)) = 0. If two rows are labelled by the same pair (α, µ), it does not matter
which one comes first.
We compute now a row echelon form of the symbol matrix. We denote the number of rows
where the pivot is of class k by β
(k)
q , the indices of the symbol Nq, and associate with each
such row its multiplicative variables x1, . . . , xk. Prolonging each equation only with respect to
its multiplicative variables yields independent equations of order q + 1, as each has a different
leading term.
Definition 6. If prolongation with respect to the non-multiplicative variables does not lead to
additional independent equations of order q + 1, in other words if
rank Mq+1 =
n∑
k=1
kβ(k)
q , (7)
Existence and Construction of Vessiot Connections 7
then the symbolNq is involutive. The differential equationRq is called involutive, if it is formally
integrable and its symbol is involutive.
The criterion (7) is also known as Cartan’s test, as it is analogous to a similar test in the
Cartan–Kähler theory of exterior differential systems. We stress again that it is valid only in
δ-regular coordinates (in fact, in other coordinate systems it will always fail).
4 The Cartan normal form
For notational simplicity, we will consider in our subsequent analysis almost exclusively first-
order equations R1 ⊆ J1π. At least from a theoretical standpoint, this is not a restriction, as any
higher-order differential equation Rq can be transformed into an equivalent first-order one (see
for example [23, Appendix A.3]). For these we now introduce a convenient local representation.
Definition 7. For a first-order differential equationR1 the following local representation, a spe-
cial kind of solved form,
uα
n = φα
n
(
x, uβ, uγ
j , uδ
n
)
1 ≤ α ≤ β
(n)
1 ,
1 ≤ j < n,
β
(n)
1 < δ ≤ m,
(8a)
uα
n−1 = φα
n−1
(
x, uβ, uγ
j , uδ
n−1
)
1 ≤ α ≤ β
(n−1)
1 ,
1 ≤ j < n− 1,
β
(n−1)
1 < δ ≤ m,
(8b)
· · · · · · · · · · · · · · · · · · · · · · · · · · ·
uα
1 = φα
1
(
x, uβ, uδ
1
) {
1 ≤ α ≤ β
(1)
1 ,
β
(1)
1 < δ ≤ m,
(8c)
uα = φα
(
x, uβ
) {
1 ≤ α ≤ β0,
β0 < β ≤ m,
(8d)
is called its Cartan normal form. The equations of zeroth order, uα = φα(x, uβ), are called
algebraic. The functions φα
k are called the right sides of R1. (If, for some 1 ≤ k ≤ n, the number
of equations is β
(k)
1 = m, then the condition β
(k)
1 < δ ≤ m is empty and no terms uδ
k appear on
the right sides of those equations.)
Here, each equation is solved for a principal derivative of maximal class k in such a way
that the corresponding right side of the equation may depend on an arbitrary subset of the
independent variables, an arbitrary subset of the dependent variables uβ with 1 ≤ β ≤ β0, those
derivatives uγ
j for all 1 ≤ γ ≤ m which are of a class j < k and those derivatives which are of the
same class k but are not principal derivatives. Note that a principle derivative uα
k may depend
on another principle derivative uγ
l as long as l < k. The equations are grouped according to
their class in descending order.
Theorem 1 (Cartan–Kähler). Let the involutive differential equation R1 be locally repre-
sented in δ-regular coordinates by the system (8a), (8b), (8c). Assume that the following initial
conditions are given:
uα(x1, . . . , xn) = fα(x1, . . . , xn), β
(n)
1 < α ≤ m; (9a)
uα(x1, . . . , xn−1, 0) = fα(x1, . . . , xn−1), β
(n−1)
1 < α ≤ β
(n)
1 ; (9b)
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·
8 D. Fesser and W.M. Seiler
uα(x1, 0, . . . , 0) = fα(x1), β
(1)
1 < α ≤ β
(2)
1 ; (9c)
uα(0, . . . , 0) = fα, 1 ≤ α ≤ β
(1)
1 . (9d)
If the functions φα
k and fα are (real-)analytic at the origin, then this system has one and only
one solution that is analytic at the origin and satisfies the initial conditions (9).
Proof. For the proof, see [19] or [23, Section 9.4] and references therein. The strategy is to
split the system into subsystems according to the classes of the equations in it (see below). The
solution is constructed step by step; each step renders a normal system in fewer independent
variables to which the Cauchy–Kovalevskaya theorem is applied. Finally, the condition that R1
is involutive leads to further normal systems ensuring that the constructed functions are indeed
solutions of the full system with respect to all independent variables. �
Under some mild regularity assumptions the algebraic equations can always be solved locally.
From now on, we will assume that any present algebraic equation has been explicitly solved,
reducing thus the number of dependent variables. We simplify the Cartan normal form of
a differential equation as given in Definition 7 into the reduced Cartan normal form. It arises by
solving each equation for a derivative uα
j , the principal derivative, and eliminating this derivative
from all other equations. Again, the principal derivatives are chosen in such a manner that their
classes are as great as possible. Now no principal derivative appears on a right side of an equation
(whereas this was possible with the non-reduced Cartan normal form of Definition 7). All the
remaining, non-principal, derivatives are called parametric. Ordering the obtained equations by
their class, we again can decompose them into subsystems:
uα
k = φα
k
(
x,u, uγ
j
)
1 ≤ j ≤ k ≤ n,
1 ≤ α ≤ β
(k)
1 ,
β
(j)
1 < γ ≤ m.
(10)
Note that the values β
(k)
1 are exactly those appearing in the Cartan test (7), as the symbol
matrix of a differential equation in Cartan normal form is automatically triangular with the
principal derivatives as pivots.
Definition 8. The Cartan characters of R1 are defined as α
(k)
1 = m− β
(k)
1 and thus equal the
number of parametric derivatives of class k and order 1.
Provided that δ-regular coordinates are chosen, it is possible to perform a closed form invo-
lution analysis for a differential equation R1 in reduced Cartan normal form. We remark that
an effective test of involution proceeds as follows (see for example [23, Remark 7.2.10]). Each
equation in (10) is prolonged with respect to each of its non-multiplicative variables. The arising
second-order equations are simplified modulo the original system and the prolongations with
respect to the multiplicative variables. The symbol N1 is involutive if, and only if, after the
simplification none of the prolonged equations is of second-order any more. The equation R1 is
involutive if, and only if, all new equations even simplify to zero, as any remaining first-order
equation would be an integrability condition.
In order to apply this test, we now prove two helpful lemmata. We introduce the set B :={
(α, i) ∈ N×N : uα
i is a principal derivative
}
, and for each (α, i) ∈ B we define Φα
i := uα
i −φα
i .
Using the contact fields (1), any prolongation of some Φα
i can be expressed in the following form.
Lemma 1. Let the differential equation R1 be represented in the reduced Cartan normal form
given by equation (10). Then for any (α, i) ∈ B and 1 ≤ j ≤ n, we have
DjΦ
α
i = uα
ij − C
(1)
j (φα
i )−
i∑
h=1
m∑
γ=β
(h)
1 +1
uγ
hjC
h
γ (φα
i ). (11)
Existence and Construction of Vessiot Connections 9
Proof. By straightforward calculation; see [6, Lemma 2.5.5]. �
For j > i, the prolongation DjΦ
α
i is non-multiplicative, otherwise it is multiplicative. Now
let j > i, so that equation (11) shows a non-multiplicative prolongation, and assume that we are
using δ-regular coordinates. According to our test, the symbol N1 is involutive if, and only if,
it is possible to eliminate on the right hand side of (11) all second-order derivatives by adding
multiplicative prolongations.
If the differential equation is not involutive, then the difference
DjΦ
α
i −DiΦ
α
j +
i∑
h=1
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
h)DhΦγ
j
does not necessarily vanish but may yield an obstruction to involution for any (α, i) ∈ B and
any i < j ≤ n. The next lemma gives all these obstructions to involution for a first-order system
in reduced Cartan normal form.
Lemma 2. Assume that δ-regular coordinates are used. For an equation in Cartan normal form
and indices i < j and α such that (α, i) ∈ B, we have the equality
DjΦ
α
i −DiΦ
α
j +
i∑
h=1
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )DhΦγ
j
= C
(1)
i (φα
j )− C
(1)
j (φα
i )−
i∑
h=1
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )C(1)
h (φγ
j ) (12)
−
i−1∑
h=1
m∑
δ=β
(h)
1 +1
uδ
hh
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ch
δ (φγ
j )
(13)
−
∑
1≤h<k<i
β
(k)
1∑
δ=β
(h)
1 +1
uδ
hk
β
(j)
1∑
γ=β
(k)
1 +1
Ck
γ (φα
i )Ch
δ (φγ
j )
(14a)
+
m∑
δ=β
(k)
1 +1
uδ
hk
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ck
δ (φγ
j ) +
β
(j)
1∑
γ=β
(k)
1 +1
Ck
γ (φα
i )Ch
δ (φγ
j )
(14b)
−
i−1∑
h=1
β
(i)
1∑
δ=β
(h)
1 +1
uδ
hi
−Ch
δ (φα
j ) +
β
(j)
1∑
γ=β
(i)
1 +1
Ci
γ(φα
i )Ch
δ (φγ
j )
(15a)
+
m∑
δ=β
(i)
1 +1
uδ
hi
−Ch
δ (φα
j ) +
β
(j)
1∑
γ=β
(i)
1 +1
Ci
γ(φα
i )Ch
δ (φγ
j ) +
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ci
δ(φ
γ
j )
(15b)
−
i−1∑
h=1
i+1≤k<j
m∑
δ=β
(k)
1 +1
uδ
hk
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ck
δ (φγ
j )
(16)
10 D. Fesser and W.M. Seiler
−
j−1∑
k=i
m∑
δ=β
(k)
1 +1
uδ
ik
−Ck
δ (φα
j ) +
β
(j)
1∑
γ=β
(i)
1 +1
Ci
γ(φα
i )Ck
δ (φγ
j )
(17)
−
i−1∑
h=1
m∑
δ=β
(j)
1 +1
uδ
hj
Ch
δ (φα
i ) +
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Cj
δ (φ
γ
j )
(18)
−
m∑
δ=β
(j)
1 +1
uδ
ij
Ci
δ(φ
α
i )− Cj
δ (φ
α
j ) +
β
(j)
1∑
γ=β
(i)
1 +1
Ci
γ(φα
i )Cj
δ (φ
γ
j )
. (19)
Proof. By a tedious but fairly straightforward computation, see [6, pages 48–55]. �
In line (12) we have collected all terms which are of lower than second order. Furthermore,
none of the appearing second-order derivatives is of a form that it could be eliminated by
adding some multiplicative prolongation. Hence, under our assumption of δ-regular coordinates,
the symbol N1 is involutive if, and only if, all the expressions in square brackets vanish. The
differential equationR1 is involutive if, and only if, in addition line (12) vanishes, as it represents
an integrability condition. Thus Lemma 2 gives us all obstructions to involution for R1 in
explicit form. They will reappear in the proof of the existence theorem for integral distributions
in Section 7.
5 The Vessiot distribution
By Proposition 1, the tangent spaces Tρ(im jqσ) of prolonged sections at points ρ ∈ Jqπ
are always subspaces of the contact distribution Cq|ρ. If the section σ is a solution of the
differential equation Rq, then by definition it furthermore satisfies im jqσ ⊆ Rq, and therefore
T (im jqσ) ⊆ TRq. Hence, the following construction suggests itself.
Definition 9. The Vessiot distribution of a differential equation Rq ⊆ Jqπ is the distribution
V[Rq] ⊆ TRq defined by
Tι
(
V[Rq]
)
= Tι
(
TRq
)
∩ Cq|Rq .
Note that the Vessiot distribution is not necessarily of constant rank along Rq (just like
the symbol Nq); for simplicity, we will almost always assume here that this is the case. This
definition of the Vessiot distribution is not the one usually found in the literature. But the
equivalence to the standard approach is an elementary exercise in computing with pull-backs.
Proposition 3. The Vessiot distribution satisfies V[Rq] = (ι∗C0
q )0.
For a differential equation given in explicitly solved form, the inclusion map ι : Rq → Jqπ is
available in closed form and can be used to calculate the pull-back of the contact forms. This
has the advantage of keeping the calculations within a space of smaller dimension, namely the
submanifold Rq. Thereby regarding the differential equation as a manifold in its own right, we
bar its coordinates to distinguish them from those of the jet bundle.
Example 1. Consider the first-order system given by the representation R1 : ut−v = vt−wx =
ux−w = 0. Then from the prolongations uxt− vx = 0 and uxt−wt = 0 follows the integrability
condition wt = vx by elimination of the second-order derivative uxt. Thus, the first projection
of R1 is (in Cartan normal form) represented by
R(1)
1 :
{
ut = v, vt = wx, wt = vx,
ux = w.
Existence and Construction of Vessiot Connections 11
It is not difficult to verify that the projected equation R(1)
1 is involutive. For coordinates on J1π
choose x, t; u, v, w; ux, vx, wx, ut, vt, wt. Since R(1)
1 is represented by a system in solved form,
it is natural to choose appropriate local coordinates for R(1)
1 , which we bar to distinguish them:
x, t; u, v, w; vx, wx. The contact codistribution for J1π is generated by
ω1 = du− uxdx− utdt, ω2 = dv − vxdx− vtdt, ω3 = dw − wxdx− wtdt.
The tangent space TR(1)
1 is spanned by ∂x, ∂t, ∂u, ∂v, ∂w, ∂vx , ∂wx and Tι(TR(1)
1 ) therefore by
the fields ∂x, ∂t, ∂u, ∂v + ∂ut , ∂w + ∂ux , ∂vx + ∂wt and ∂wx + ∂vt . This space is annihilated by
ω4 = dux − dw, ω5 = dvx − dwt, ω6 = dwx − dvt, ω7 = dut − dv.
These seven one-forms ω1, . . . , ω7 annihilate the Vessiot distribution V[R(1)
1 ], which is spanned
by the four vector fields
X1 = ∂x + ux∂u + vx∂v + wx∂w + vx∂ut + wx∂ux , X3 = ∂vx + ∂wt ,
X2 = ∂t + ut∂u + vt∂v + wt∂w + vt∂ut + wt∂ux , X4 = ∂vt + ∂wx .
In local coordinates on R(1)
1 , these four vector fields become
X̄1 = ∂x + w∂u + vx∂v + wx∂w, X̄3 = ∂vx ,
X̄2 = ∂t + v∂u + wx∂v + vx∂w, X̄4 = ∂wx .
They satisfy ι∗X̄i = Xi, as a simple calculation using the Jacobian matrix for Tι shows. The
vector fields X̄i are annihilated by the pull-backs of the contact forms, ι∗ω1 = du−uxdx−utdt,
ι∗ω2 = dv − xxdx − xtdt and ι∗ω3 = dw − wxdx − wtdt (the pullbacks of the remaining four
one-forms ω4, . . . , ω7 trivially vanish on R(1)
1 ).
For a fully nonlinear differential equation Rq, in particular an implicit one, this approach
to compute its Vessiot distribution V[Rq] via a pull-back is in general not effectively feasible.
However, applying directly our definition of V[Rq], it is easily possible even for such equations to
determine effectively Tι(V[Rq]), in other words: to realize it as a subbundle of T (Jqπ)|Rq . The
contact fields (1) form a basis for Cq. It follows that for any vector field X̄ ∈ V[Rq], coefficients
ai, bα
µ ∈ F(Rq), where 1 ≤ i ≤ n, 1 ≤ α ≤ m and |µ| = q, exist such that
ι∗X̄ = aiC
(q)
i + bα
µCµ
α . (20)
If the differential equation Rq is locally represented by Φτ = 0, where 1 ≤ τ ≤ t, it follows from
the tangency of the vector fields in V[Rq] that dΦτ (ι∗X̄) = ι∗X̄(Φτ ) = 0 and thus the coefficient
functions must satisfy the following system of linear equations:
C
(q)
i (Φτ )ai + Cµ
α(Φτ )bα
µ = 0, (21)
where 1 ≤ τ ≤ t. Note that this approach to determine the Vessiot distribution is closely re-
lated to prolonging the differential equation Rq and requires essentially the same computations.
Indeed, the formal derivative (5) can be written in the form
DiΦτ = C
(q)
i (Φτ ) + Cµ
α(Φτ )uα
µ+1i
= 0 (22)
and in the context of the order-by-order determination of formal power series solutions (see
for example [23, Section 2.3]) these equations are considered an inhomogeneous system for the
Taylor coefficients of order q + 1 depending on the lower order coefficients. Taking this point
of view, we may call (21) the “projective” version of (22). In fact for n = 1, that is, for ordinary
differential equations, this is even true in a rigorous sense.
12 D. Fesser and W.M. Seiler
Example 2. We consider the fully nonlinear first-order ordinary differential equationR1 locally
defined by (u′)2 + u2 + x2 = 1. The contact distribution C1 is spanned by the two vector fields
X1 = ∂x + u′∂u and X2 = ∂u′ and the Vessiot distribution Tι(V[Rq]) consists of all linear
combinations of these two fields which are tangent to R1. Setting ω = xdx + udu + u′du′, we
thus have to solve the linear equation ω(aX1 + bX2) = 0 in order to determine Tι(V[Rq]). Its
solution requires a case distinction (which is typical for fully nonlinear differential equations).
If u′ 6= 0, then we find
Tι(V[Rq]) = 〈u′∂x + (u′)2∂u − (x + u′u)∂u′〉.
For u′ = 0 and x 6= 0, the Vessiot distribution is spanned by the vertical contact field X2.
Finally, for x = u′ = 0 the rank of the Vessiot distribution jumps, as at these points the whole
contact plane is contained in it.
The definition of the symbol Nq and of the Vessiot distribution V[Rq], respectively, of a diffe-
rential equation Rq ⊆ Jqπ immediately imply the following generalisation of the above discussed
splitting of the contact distribution Cq = V πq
q−1⊕H (such a splitting of the Vessiot distribution
is also discussed by Lychagin and Kruglikov [14, 15] where the Vessiot distribution is called
“Cartan distribution”).
Proposition 4. For any differential equation Rq, its symbol is contained in the Vessiot distri-
bution: Nq ⊆ V[Rq]. The Vessiot distribution can therefore be decomposed into a direct sum
V[Rq] = Nq ⊕H (23)
for some complement H (which is not unique).
Such a complement H is necessarily transversal to the fibration Rq → X and thus leads
naturally to connections: provided dimH = n, it may be considered the horizontal bundle of
a connection of the fibred manifold Rq → X .
Definition 10. Any such connection is called a Vessiot connection for Rq.
In general, the Vessiot distribution V[Rq] is not involutive (that is, closed under the Lie
bracket; an exception are formally integrable equations of finite type [23, Remark 9.5.8]), but it
may contain involutive subdistributions. If these are furthermore transversal (to the fibration
Rq → X ) and of dimension n, then they define a flat Vessiot connection.
Lemma 3. If the section σ : X → E is a solution of the equation Rq, then the tangent bundle
T (im jqσ) is an n-dimensional transversal involutive subdistribution of V[Rq]|im jqσ. Conversely,
if U ⊆ V[Rq] is an n-dimensional transversal involutive subdistribution, then any integral man-
ifold of U has locally the form im jqσ for a solution σ of Rq.
Proof. Let σ be a local solution of the equation Rq. Then it satisfies by Definition 3 im jqσ ⊆
Rq and thus T (im jqσ) ⊆ TRq. Besides, by the definition of the contact distribution, for x ∈ X
with jqσ(x) = ρ ∈ Jqπ, the tangent space Tρ(im jqσ) is a subspace of Cq|ρ. By definition of the
Vessiot distribution, it follows Tρ(im jqσ) ⊆ Tι(TρRq) ∩ Cq|ρ, which proves the first claim.
Now let U ⊆ V[Rq] be an n-dimensional transversal involutive subdistribution. Then ac-
cording to the Frobenius theorem, U has n-dimensional integral manifolds. By definition,
Tι(V[Rq]) ⊆ Cq|Rq ; this characterises prolonged sections. Hence, for any integral manifold of U
there is a local section σ such that the integral manifold is of the form im jqσ. Furthermore, the
integral manifold is a subset of Rq. Thus it corresponds to a local solution of Rq. �
This simple observation forms the basis of Vessiot’s approach to the analysis of Rq: he
proposed to construct all flat Vessiot connections. Before we do this, we first show how integral
elements appear in this program.
Existence and Construction of Vessiot Connections 13
Proposition 5. Let U ⊆ V[Rq] be a transversal subdistribution of the Vessiot distribution of
constant rank k. Then the spaces Uρ are k-dimensional integral elements for all points ρ ∈ Rq
if, and only if, [U ,U ] ⊆ V[Rq].
Proof. Let {ω1, . . . , ωr} be a basis of the codistribution ι∗C0
q . Then an algebraic basis of the
ideal I[Rq] is {ω1, . . . , ωr, dω1, . . . , dωr}. Any vector field X ∈ U trivially satisfies ωi(X) = 0
by Proposition 3. For arbitrary fields X1, X2 ∈ U , we have
dωi(X1, X2) = X1
(
ωi(X2)
)
−X2
(
ωi(X1)
)
+ ωi
(
[X1, X2]
)
.
The first two summands on the right hand side vanish trivially and the remaining equation
implies our claim. �
We call a subdistribution U ⊆ V[Rq] satisfying the conditions of Proposition 5 an integral
distribution for the differential equation Rq. In the literature [24], the terminology “involution”
is common for such distributions which, however, is confusing. Note that generally an integral
distribution is not integrable; the name only reflects the fact that it consists of integral elements.
A general differential equations Rq does not necessarily possess any Vessiot connection (not
even a non-flat one). Their existence is linked to the absence of integrability conditions. More
precisely, we obtain the following characterisation.
Proposition 6. Let Rq be a differential equation. Then its Vessiot distribution possesses locally
a direct decomposition with an n-dimensional complement H such that V[Rq] = Nq ⊕H if, and
only if, there are no integrability conditions which arise as the prolongation of equations of lower
order in the system.
Proof. Let Rq be locally represented by
Rq :
{
Φτ
(
x,u(q)
)
,
Ψσ
(
x,u(q−1)
)
,
such that the equations Φτ (x,u(q)) = 0 do not imply lower-order equations which are indepen-
dent of the equations Ψσ(x,u(q−1)) = 0. Let u(q) denote the subset of all derivatives of order q
only; then the Jacobi matrix
(
∂Φτ (x,u(q))/∂u(q)
)
has maximal rank. If we proceed as in the
last proof, then the ansatz (20) for the determination of the Vessiot distribution yields for the
above representation the linear system
C
(q)
i (Φτ )ai + Cµ
α(Φτ )bα
µ = 0, (24)
C
(q)
i (Ψσ)ai = 0.
Here, the matrix Cµ
α(Φτ ) has maximal rank, too; thus the equations C
(q)
i (Φτ )ai + Cµ
α(Φτ )bα
µ = 0
can be solved for a subset of the unknowns bα
µ. And since no terms of order q are present in
Ψσ(x,u(q−1)) = 0, we have C
(q)
i (Ψσ) = DiΨ
σ. Recall that we consider the Vessiot distribution,
and thus the linear system (24), only on Rq. It follows that the subsystem C
(q)
i (Ψσ)ai = 0
becomes trivial if, and only if, no integrability conditions arise from the prolongation of lower
order equations. And if, and only if, this is the case, then (24) has for each 1 ≤ j ≤ n a solution
where aj = 1 while all other ai are zero. The existence of such a solution is equivalent to the
existence of an n-dimensional transversal complement H. �
From the proof of this proposition follows that for the determination of the Vessiot distri-
bution V[Rq], equations of order less than q in the local representation of Rq can be ignored
if there are no integrability conditions which arise from equations of lower order. It is the
14 D. Fesser and W.M. Seiler
integrability conditions which arise as prolongations of lower order equations that hinder the
construction of n-dimensional complements, while those which follow from the relations between
cross derivatives do not influence this approach.
Remark 1. If one does not care about the distinction between different kinds of integrability
conditions and simply requires that Rq = R(1)
q (meaning that no integrability conditions at all
appear in the first prolongation of Rq), then one can provide a more geometric proof for the
existence of an n-dimensional complement (of course, in contrast to Proposition 6, the converse
is not true then).
The assumption Rq = R(1)
q implies that to every point ρ ∈ Rq at least one point ρ̂ ∈ Rq+1
with πq+1
q (ρ̂) = ρ exists. We choose such a point and consider im Γq+1(ρ̂) ⊂ Tρ(Jqπ). By
definition of the contact map Γq+1, this is an n-dimensional transversal subset of Cq|ρ. Thus
there only remains to show that it is also tangential to Rq, as then we can define a complement
by Tρι(Hρ) = im Γq+1(ρ̂). But this tangency is a trivial consequence of ρ̂ ∈ Rq+1; using for
example the local coordinates expression (3) for the contact map and a local representation
Φτ = 0 of Rq, one immediately sees that the vector vi = Γq+1(ρ̂, ∂xi) ∈ Tρ(Jqπ) satisfies
dΦτ |ρ(vi) = DiΦτ (ρ̂) = 0 and thus is tangential to Rq.
Hence it is possible to construct for each point ρ ∈ Rq a transversal complement Hρ such that
Vρ[Rq] = (Nq)ρ⊕Hρ. There remains to show that these complements can be chosen so that they
form a smooth distribution. Our assumption Rq = R(1)
q implies that the restricted projection
π̂q+1
q : Rq+1 → Rq is a surjective submersion, that is, it defines a fibred manifold. Thus if we
choose a section γ : Rq → Rq+1 and then always take ρ̂ = γ(ρ), it follows immediately that the
corresponding complements Hρ define a smooth distribution as required.
Example 3. Consider again the differential equation R1 in Example 1. It is locally represented
by the same equations as R(1)
1 , except that the integrability condition wt = vx is missing. The
matrix of Tι for the system R1 has eleven rows and eight columns—one column more than
the symbol matrix for the system R(1)
1 . The symbol Tι(N1) of R1 is 3-dimensional, spanned
by ∂vx , ∂wt and ∂wx + ∂vt , while the symbol Tι(N (1)
1 ) of R(1)
1 has dimension 2 and is spanned
by ∂vx + ∂wt and ∂wx + ∂vt . But the one-forms ω1, ω2 and ω3 (and their pull-backs ι∗ω1 =
du − uxdx − utdt, ι∗ω2 = dv − vxdx − vtdt and ι∗ω3 = dw − wxdx − wtdt) are the same, and
therefore the coordinate expressions for the vector fields X1 and X2 (and their representations
X̄1 = ∂x + ux∂u + vx∂v + wx∂w and X̄2 = ∂t + ut∂u + vt∂v + wt∂w in TR1 and TR(1)
1 ) look alike
(see Example 1 for their representations). The integrability condition wt = vx does not influence
the results as it stems from the equality of the cross derivatives, utx = vx and uxt = wt, and not
from the prolongation of a lower order equation.
Now consider for comparison the differential equation which is locally represented by
R̃1 :
ut = v, vt = wx, wt = vx,
ux = w,
u = x.
It arises from the system R(1)
1 in Example 1 by adding the algebraic equation u = x. Proceeding
as in Example 1, we find that the Vessiot distribution V[R̃1] is spanned by the three vector
fields
∂vx , ∂wx and v∂x + (1− w)∂t +
(
wx(1− w) + vvx
)
∂v +
(
vx(1− w) + vwx
)
∂w.
The vector fields ∂vx and ∂wx generate the symbol Ñ1; any of its complements in V[R̃1] is one-
dimensional and, as the dimension of the base space is two, none of them has the right dimension
to be the horizontal space of a connection.
Existence and Construction of Vessiot Connections 15
The reason for this is that R̃1 is not formally integrable, as the prolongation of the algebraic
equation u = x leads to the integrability conditions ux = 1 and ut = 0. Projecting the prolonged
equation gives
R̃(1)
1 :
ut = v = vt = wx = wt = vx = 0,
ux = w = 1,
u = x.
Now the symbol vanishes, and so do the pull-backs of the contact forms: ι∗ω1 = dx − dx = 0,
ι∗ω2 = dv = 0, ι∗ω3 = dw = 0. Therefore we find V[R̃(1)
1 ] = Ñ (1)
1 ⊕H = {0} ⊕ 〈∂x, ∂t〉. As the
Lie brackets of ∂x and ∂t trivially vanish, T R̃(1)
1 = V[R̃(1)
1 ] = H = 〈∂x, ∂t〉 is a two-dimensional
involutive distribution.
Any n-dimensional complement H is obviously a transversal subdistribution of V[Rq], but not
necessarily involutive. Conversely, any n-dimensional subdistribution H of V[Rq] is a possible
choice as a complement. Choosing a “reference” complement H0 with a basis (Xi : 1 ≤ i ≤ n),
a basis for any other complement H arises by adding some vertical fields to the vectors Xi.
We will follow this approach in the next section. For the remainder of this section we turn our
attention to the choice of a convenient basis of V[Rq] that will facilitate our computations.
Let r := dimNq. As an intersection of two involutive distributions, the symbol Nq is an
involutive distribution, too. Hence, there exists a basis (Yk : 1 ≤ k ≤ r) for it whose Lie brackets
vanish: [Yk, Y`] = 0 for all 1 ≤ k, ` ≤ r. Since the vertical bundle V πq
q−1 is also involutive, we
can decompose it into
V πq
q−1 = Nq ⊕W,
where W is again an involutive distribution. It can be spanned by vector fields W1,W2, . . . ,Ws
(where s =
∑n
k=1 β
(k)
q equals the number of principal derivatives) which are chosen such that
we have [Wa,Wb] = 0 for all 1 ≤ a, b ≤ s. In local coordinates, a particularly convenient choice
for the fields Yk and Wa exists. We first choose for any 1 ≤ k ≤ r a parametric derivative uα
µ,
that is (α, µ) /∈ B, and set Yk := Y α
µ := ι∗(∂uα
µ
); then we choose for any 1 ≤ a ≤ s a principal
derivative uα
µ, that is (α, µ) ∈ B, and set Wa := Wα
µ := ∂uα
µ
.
The reference complementH0 is chosen as follows. Any basis of it must consist of n transversal
contact fields. Since the fields Cµ
α are vertical, we can always use a basis (X̃1 : 1 ≤ i ≤ n) of the
form
X̃i = C
(q)
i + ξα
iµCµ
α
with some coefficient functions ξα
iµ chosen such that X̃i is tangential to Rq. The fields Cµ
α also
span the vertical bundle V πq
q−1 and hence we may exploit the above decomposition for a further
simplification of the basis. By subtracting from each X̃i a suitable linear combination of the
fields Yk spanning the symbol Nq, we arrive at a basis (Xi : 1 ≤ i ≤ n) where
Xi = C
(q)
i + ξa
i Wa. (25)
As already mentioned above, the Vessiot distribution V[Rq] is not necessarily involutive. So
it is not surprising that its structure equations are going to be important later. We may extend
the above chosen basis (Xi, Yk) of V[Rq] to a basis of the derived Vessiot distribution, V ′[Rq],
by adding vector fields Zc, 1 ≤ c ≤ C := dimV ′[Rq] − dimV[Rq], where, using (2), for each c
we have some coefficients κα
cν such that Zc = κα
cν∂uα
ν
with |ν| = q − 1. By construction, the
non-vanishing structure equations of V[Rq] take now the form
[Xi, Xj ] = Θc
ijZc and [Xi, Yk] = Ξc
ikZc (26)
16 D. Fesser and W.M. Seiler
for 1 ≤ i, j ≤ n and 1 ≤ k ≤ r, with smooth functions Θc
ij and Ξc
ik. (For the complete set of
structure equations, we have to add [Yk, Yl] = 0 for 1 ≤ k, l ≤ r.)
Remark 2. Since the vector fields Zc, which appear on the right sides of the structure equa-
tions (26), may, for q = 1, span a proper subspace in 〈∂uα : 1 ≤ α ≤ m〉, about the exact size
of which we know nothing, we write them as linear combinations Zc =: κα
c ∂uα . The structure
equations then become
[Xi, Xj ] = Θc
ijκ
α
c ∂uα =: Θα
ij∂uα , 1 ≤ i, j ≤ n, (26a′)
[Xi, Yk] = Ξc
ikκ
α
c ∂uα =: Ξα
ik∂uα , 1 ≤ i ≤ n, 1 ≤ k ≤ r. (26b′)
Knowing the larger sets of coefficients Θα
ij , Ξα
ik, we can reconstruct the true structure
coefficients Θc
ij , Ξc
ik by solving the overdetermined systems of linear equations
Θc
ijκ
α
c = Θα
ij and Ξc
ikκ
α
c = Ξα
ik.
This is always possible since the fields Zc are assumed to be part of a basis for the derived
Vessiot distribution V ′[R1] and therefore linearly independent. Thus there exist some coefficient
functions κc
α such that
Θc
ij = Θα
ijκ
c
α and Ξc
ik = Ξα
ikκ
c
α.
For our later proof of an existence theorem for integral distributions, we will have to analyse
certain matrices with the coefficients Θc
ij and Ξc
ik as their entries. It turns out that this analysis
becomes simpler, if we use the extended sets of coefficients Θα
ij and Ξα
ik instead.
For a first-order equation R1 with Cartan normal form (8) satisfying the assumptions of
Proposition 6 it is possible to perform this process explicitly. We choose as a reference comple-
ment H0 the linear span of the vector fields
ι∗X̄i = C
(q)
i +
∑
(α,µ)∈B
C
(q)
i (φα
µ)Cµ
α (27)
for 1 ≤ i ≤ n. One verifies in a straightforward computation that (27) represents a valid choice
(see [6, Proposition 3.1.19]). Using this reference complement, we can explicitly evaluate the Lie
brackets (26) on R1. As we are not able to determine a simple expression for the derived Vessiot
distribution, we follow the approach of Remark 2 and consider the equations (26′) instead.
Lemma 4. Let i < j, without loss of generality. Then we obtain for the extended set of structure
coefficients Θα
ij in local coordinates on R1 the following results:
Θα
ij =
C
(1)
i (φα
j )− C
(1)
j (φα
i ), (α, i) ∈ B and (α, j) ∈ B,
C
(1)
i (φα
j ), (α, i) 6∈ B and (α, j) ∈ B,
0, (α, i) 6∈ B and (α, j) 6∈ B.
(28)
Proof. By a rather straightforward computation; see [6, pages 75 and 76]. �
We collect these coefficients into vectors Θij which have m rows each where the entries are
ordered according to increasing α. It is useful to denote the symbol fields Yk = ι∗(∂
uβ
h
) by using
double indices: Yk = Y β
h for any (β, h) 6∈ B.
Existence and Construction of Vessiot Connections 17
Lemma 5. Set Ȳk =: Ȳ β
h , then the extended set of structure coefficients Ξα
ik in local coordinates
on R1 are
Ξα
ik =
−Ch
β (φα
i ), (α, i) ∈ B,
−1, (α, i) /∈ B and (α, i) = (β, h),
0, (α, i) /∈ B and (α, i) 6= (β, h).
(29)
Proof. Again by a rather straightforward computation; see [6, page 76]. �
Example 4. We calculate the structure equations for Example 1. Here, β
(1)
1 = 1 and β
(2)
1 = 3.
We set X̄3 =: Ȳ1 = Ȳ 2
1 and X̄4 =: Ȳ2 = Ȳ 3
1 . Then besides the trivial structure equations we get:
[X̄1, X̄2] = 0, [X̄1, Ȳ1] = [X̄2, Ȳ2] = −∂v, [X̄2, Ȳ1] = [X̄1, Ȳ2] = −∂w.
Noting that here the set B contains the pairs (1, 1) ≡ ux, (1, 2) ≡ ut, (2, 2) ≡ vt and (3, 2) ≡ wt
while (2, 1) ≡ vx and (3, 1) ≡ wx are not in B, one easily verifies that all coefficients are as
given in equations (28) and (29).
We end this section with two technical remarks on how these coefficients are collected into
matrices Ξi. The examination of the ranks of these matrices Ξi is basic for the proof of the
existence Theorem 2 for integral distributions.
Remark 3. Some of the terms −Ch
β (φα
i ) where (α, i) ∈ B vanish, too: all of the parametric
derivatives on the right side of an equation Φα
i in the reduced Cartan normal form (10) are of
a class lower than that of the equation’s left side as otherwise we would solve this equation for
the derivative of highest class. This means −Ch
β (φα
i ) = 0 whenever i = cls(uα
i ) < cls(uβ
h) = h.
We collect the coefficients Ξα
ik into matrices Ξi using i as the number of the matrix to which
the entry Ξα
ik belongs, α as the row index of the entry and k as its column index. These matrices
have m rows each, ordered according to increasing α; and they have r = dimN1 columns each
of which can be labelled by pairs (β, h) 6∈ B or the symbol fields Ȳk = Ȳ β
h . More precisely, for
1 ≤ h ≤ n, we set
−Ch
β
(h)
1 +1
(φ1
i ) −Ch
β
(h)
1 +2
(φ1
i ) · · · −Ch
m(φ1
i )
−Ch
β
(h)
1 +1
(φ2
i ) −Ch
β
(h)
1 +2
(φ2
i ) · · · −Ch
m(φ2
i )
...
...
. . .
...
−Ch
β
(h)
1 +1
(φβ
(i)
1
i ) −Ch
β
(h)
1 +2
(φβ
(i)
1
i ) · · · −Ch
m(φβ
(i)
1
i )
=: [Ξi]h. (30)
Such a matrix with an upper index h collects all those Ξα
ik into a block where (α, i) ∈ B. For
any 1 ≤ i ≤ n, we have m− β
(i)
1 = α
(i)
1 , so such a matrix has β
(i)
1 rows and α
(i)
1 columns. Since,
for any h with i < h and for all β
(h)
1 + 1 ≤ β ≤ m, we have −Ch
β (φα
i ) = 0, such matrices [Ξi]h
where i < h are zero. Furthermore, for 1 ≤ h ≤ n, we assemble the remaining terms Ξα
ik (which
are those where (α, i) /∈ B) in a matrix. As above, let Ȳk = Ȳ β
h , and denote any entry Ξα
ik
accordingly, for the sake of introducing the following matrix, by h
βΞα
i . Now set
h
β
(h)
1 +1
Ξ
β
(i)
1 +1
i
h
β
(h)
1 +2
Ξ
β
(i)
1 +1
i · · · h
mΞ
β
(i)
1 +1
i
h
β
(h)
1 +1
Ξ
β
(i)
1 +2
i
h
β
(h)
1 +2
Ξ
β
(i)
1 +2
i · · · h
mΞ
β
(i)
1 +2
i
...
...
. . .
...
h
β
(h)
1 +1
Ξm
i
h
β
(h)
1 +2
Ξm
i · · · h
mΞm
i
=: [Ξi]h.
18 D. Fesser and W.M. Seiler
0 · · · 0
0 · · · 0
. . .
· · ·
· · ·
α
(i)
1
β
(i)
1
α
(1)
1 α
(2)
1 α
(i−1)
1 α
(i)
1 Σ
n
h=i+1α
(h)
1
Figure 1. A sketch for the matrix Ξi in equation (31).
For any 1 ≤ h ≤ n, such a matrix with the index h written below has α
(i)
1 rows and α
(h)
1
columns. Let for any natural numbers a and b denote 0a×b the a× b zero matrix. According to
equation (29), we have
[Ξi]h =
{
−1
α
(i)
1
, h = i,
0
α
(i)
1 ×α
(h)
1
, h 6= i.
Using the matrices [Ξi]h and [Ξi]h as blocks, we now build the matrix
Ξi =
(
[Ξi]1 [Ξi]2 · · · [Ξi]n
[Ξi]1 [Ξi]2 · · · [Ξi]n
)
.
Taking into account what we have noted on its entries, this means
Ξi =
(
[Ξi]1 · · · [Ξi]i−1 [Ξi]i 0 · · · 0
0 · · · 0 −1
α
(i)
1
0 · · · 0
)
. (31)
A sketch of such a matrix Ξi where the entries which may be different from zero are marked as
shaded areas and −1
α
(i)
1
as a diagonal line is given in Fig. 1.
For all h where 1 ≤ h ≤ n, we call the block [Ξi]h in Ξi stacked upon the block [Ξi]h in Ξi
the hth block of columns in Ξi. For those h with β
(h)
1 = m the hth block of columns is empty.
Now the symbol fields Ȳ β
h , or equivalently the pairs (β, h) 6∈ B, are used to order the dimN1 = r
columns of Ξi, according to increasing h into n blocks (empty for those h with α
(h)
1 = 0) and
within each block according to increasing β (with β
(j)
1 + 1 ≤ β ≤ m).
Remark 4. This means, the columns in Ξi are ordered increasingly with respect to the term-
over-position lift of the degree reverse lexicographic ranking. Therefore, the entry −Ch
β
(h)
1 +γ
(φα
i )
stands in the matrix Ξi in line α, in the hth block of columns of which it is the γth one from
the left. Entries different from zero and from −1 may appear in Ξi only in a [Ξi]h for h ≤ i.
To be exact, for any class i, the matrix Ξi has α
(i)
1 rows where all entries are zero with only one
exception: for each 1 ≤ `i < α
(i)
1 we have
Ξ
β
(i)
1 +`i
i k = −δ` k,
Existence and Construction of Vessiot Connections 19
where ` :=
∑i−1
h=1 α
(h)
1 + `i. The entries in the remaining upper β
(i)
1 rows are −Ch
β (φα
i ). The
potentially non-trivial ones of them are marked as shaded areas in Fig. 1.
Note that for a differential equation with constant coefficients all vectors Θij vanish and for
a maximally over-determined equation there are no matrices Ξi.
The unit block of α
(i)
1 rows, 1
α
(i)
1
, leads immediately to the estimate
α
(i)
1 ≤ rank Ξi ≤ min
{
m,
i∑
h=1
α
(h)
1
}
.
Example 5. For Example 1, we have n = 2 and therefore two matrices Ξ1 and Ξ2. Their
entries follow immediately from our results in Example 4. For i = 1, we have
Ξ1 =
0 0
−1 0
0 −1
.
The first line is [Ξ1]1 = (0, 0). The unit block below it is [Ξ1]1. (We have β
(1)
1 = 1 and therefore
α
(1)
1 = 2. There is only h = 1 to consider, as both parametric derivatives vx and wx are with
respect to x only.) Because of the very simple nature of our system, we find here by accident that
Ξ1 = Ξ2. However, [Ξ2]1 = Ξ2 is the whole matrix while there is no [Ξ2]1 because β
(2)
1 = m = 3
and therefore α
(2)
1 = 0. Finally, we obtain Θ12 = (vx − wt, 0, 0)t. (The t top right marks the
transpose.) Note that its first entry is in fact the integrability condition.
6 Flat Vessiot connections
In this section, we develop an approach for constructing flat Vessiot connections which improves
recent approaches [5, 24, 27]: we exploit the splitting of V[Rq] suggested by (23) to introduce
convenient bases for integral distributions which yield structure equations that are particularly
simple. Finally, we give necessary and sufficient conditions for Vessiot’s approach to succeed.
Let Rq locally be represented by the system Φτ (x,u(q)) = 0 where 1 ≤ τ ≤ t. Our goal
is the construction of all n-dimensional transversal involutive subdistributions U within the
Vessiot distribution V[Rq]. Taking for the Vessiot distribution the basis (Xi, Yk) where the
vector fields Yk are the above mentioned basis of the symbol Tι(Nq) with vanishing Lie brackets
and the vector fields Xi are given in (25), we make for the basis (Ui : 1 ≤ i ≤ n) of such
a subdistribution U the ansatz
Ui = Xi + ζk
i Yk
with yet undetermined coefficient functions ζk
i ∈ F(Rq). This ansatz follows naturally from
our considerations so far, as the fields Xi are transversal to the fibration over X and span a
reference complement to the symbol and in Nq all fields Yk are vertical.
Since the fields Ui are in triangular form, the distribution U is involutive if, and only if, their
Lie brackets vanish, and using the structure equations (26) this means:
[Ui, Uj ] = [Xi, Xj ] + ζk
i [Yk, Xj ] + ζk
j [Xi, Yk] +
(
Ui(ζk
j )− Uj(ζk
i )
)
Yk
=
(
Θc
ij −Ξc
jkζ
k
i + Ξc
ikζ
k
j
)
Zc +
(
Ui(ζk
j )− Uj(ζk
i )
)
Yk = 0. (32)
It follows from the definition of the fields Yk and Zc that they are linearly independent and so
their coefficients must vanish for U to be involutive. Thus the Lie bracket (32) yields two sets
20 D. Fesser and W.M. Seiler
of conditions for the coefficient functions ζk
i : a system of algebraic equations
Gc
ij := Θc
ij −Ξc
jkζ
k
i + Ξc
ikζ
k
j = 0,
{
1 ≤ c ≤ C,
1 ≤ i < j ≤ n,
(33)
and a system of differential equations
Hp
ij := Ui(ζ
p
j )− Uj(ζ
p
i ) = 0,
{
1 ≤ p ≤ r,
1 ≤ i < j ≤ n.
(34)
In the algebraic system (33) the true structure coefficients Θc
ij , Ξc
jk appear. For our subsequent
analysis we follow Remark 2 and replace them by the extended set of coefficients Θα
ij , Ξα
jk. This
corresponds to replacing (33) by an equivalent but larger linear system of equations which is,
however, simpler to analyse.
Obviously, (33) is an inhomogeneous linear system to which any solution method for linear
systems can be applied. We shall see in the next section that its structure (induced by the
structure equations of the Vessiot distribution) allows us to decompose it into simpler subsystems
which are considered step by step. Many papers on the Vessiot theory (see for example [5, 24])
study at this stage a quadratic system which only by such a step-by-step approach can be
reduced to a series of linear problems. The linearity of (33) is a simple consequence of our choice
of a basis for V[Rq] which in turn exploits the natural splitting V[Rq] = Nq ⊕H.
Remark 5. The vector fields Yk lie in the Vessiot distribution V[R1]. Thus, according to
Proposition 5, the subdistribution U is an integral distribution if, and only if, the coefficients ζi
k
satisfy the algebraic conditions (33). This observation permits us immediately to reduce the
number of unknowns in our ansatz. Assume that we have values 1 ≤ i, j ≤ n and 1 ≤ α ≤ m
such that both (α, i) and (α, j) are not contained in B, that is, uα
i and uα
j are both parametric
derivatives (and thus obviously the second-order derivative uα
ij , too). Then there exist two
symbol fields Yk = ι∗(∂uα
i
) and Yl = ι∗(∂uα
j
). Now it follows from the coordinate form (3) of
the contact map that the subdistribution U , spanned by the vector fields Uh = Xh + ζk
hYk for
1 ≤ h ≤ n, can be an integral distribution if, and only if,
ζk
j = ζ l
i or, equivalently, ζ
(α,i)
j = ζ
(α,j)
i . (35)
for all 1 ≤ i < j ≤ n and 1 ≤ k, l ≤ r.
As the unknowns ζk
j may be understood as labels for the columns of the matrices Ξh, this
identification leads to a contraction of these matrices. The contracted matrices, denoted by
Ξ̂h, arise as follows: whenever ζk
j = ζ l
i then the corresponding columns of Ξh are added; their
sum replaces the first of these columns, while the second column is dropped. Similarly, we
introduce reduced vectors ζ̂h where the redundant components are left out. From now on, we
always understand that in the equations above this reduction has been performed. Otherwise
some results would not be correct (see Example 10 below).
For a more thorough outline of these technical details, see [6, Lemma 3.3.5 and Remark 3.3.6].
Example 6. We consider a first-order equation in one dependent variable
R1 :
un = φn(x, u, u1, . . . , un−r−1),
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·
un−r = φn−r(x, u, u1, . . . , un−r−1).
It is easy to show that the symbol of such an equation is always involutive and that the used
coordinates are δ-regular. In order to simplify the notation, we use the following convention for
Existence and Construction of Vessiot Connections 21
the indices: 1 ≤ k, ` < n − r and n − r ≤ a, b ≤ n. Thus we may express our system in the
concise form ua = φa(x, u, uk) and local coordinates on the submanifold R1 are (x, u, uk).
The pull-back of the contact form ω = du− uidxi generating the contact codistribution C0
1 is
ι∗ω = du− ukdxk − φadxa and V[R1] = 〈X̄1, . . . , X̄n, Ȳ1, . . . , Ȳn−r−1〉 where
X̄k = ∂xk + uk∂u, X̄a = ∂xa + φa∂u, Ȳk = ∂uk
.
The fields Ȳk span the symbol N1 and the fields X̄i our choice of a reference complement H0.
Setting Z̄ = ∂u, the structure equations of V[R1] are
[X̄k, X̄`] = 0, [Ȳk, Ȳ`] = 0,
[X̄k, X̄a] = X̄k(φa)Z̄, [X̄a, X̄b] =
(
X̄a(φb)− X̄b(φa)
)
Z̄,
[X̄k, Ȳ`] = −δk`Z̄, [X̄a, Ȳk] = −Ȳk(φa)Z̄.
Now we make the above discussed ansatz Ui = Xi + ζk
i Yk for the generators of a transversal
complement H. Modulo the Vessiot distribution V[R1] we obtain for their Lie brackets
[Uk, U`] ≡ (ζ`
k − ζk
` )Z mod V[R1], (36a)
[Ua, Uk] ≡
(
ζk
a − ζ`
kY`(φa)−Xk(φa)
)
Z mod V[R1], (36b)
[Ua, Ub] ≡
(
ζk
aYk(φb)− ζ`
bY`(φa) + Xa(φb)−Xb(φa)
)
Z mod V[R1]. (36c)
The algebraic system (33) is now obtained by requiring that all the expressions in parentheses
on the right hand sides vanish. Its solution is straightforward. The first subsystem (36a) implies
the equalities ζ`
k = ζk
` . This result was to be expected by the discussion in Remark 5: both uk
and u` are parametric derivatives for R1 and thus we could have made this identification already
in our ansatz for the complement. The second subsystem (36b) yields that ζk
a = ζ`
kY`(φa) +
Xk(φa). If we enter these results into the third subsystem (36c), then all unknowns ζ drop out
and the solvability condition
Xa(φb)−Xb(φa) + Xk(φa)Yk(φb)−Xk(φb)Yk(φa) = 0
arises. Thus in this example the algebraic system (33) has a solution if, and only if, this condition
is satisfied.
Comparing with the classical theory of such systems, one easily verifies that this solvability
condition is equivalent to the vanishing of the Mayer or Jacobi bracket [ua − φa, ub − φb] on the
submanifold R1, which in turn is a necessary and sufficient condition for the formal integrability
of the differential equation R1. Thus we may conclude that R1 possesses n-dimensional integral
distributions if, and only if, it is formally integrable (which in our case is also equivalent to R1
being involutive).
Thus here involution can be decided solely on the basis of the algebraic system (33). We will
show in the next section that this observation does not represent a special property of a very
particular class of differential equations but a general feature of the Vessiot theory.
7 The existence theorem for integral distributions
Now the question arises, when the combined system (33), (34) has solutions? We begin by
analysing the algebraic part (33). We use for this analysis a step-by-step approach originally
proposed by Vessiot [28]. Our analysis will automatically reveal necessary and sufficient assump-
tions for it to succeed. As outlined in Remark 5, we replace ζ2 by ζ̂2 since for the entries ζ
(β,1)
2
where β
(2)
1 +1 ≤ β ≤ m we know already from equation (35) that ζ
(β,1)
2 = ζ
(β,2)
1 . Thus we begin
22 D. Fesser and W.M. Seiler
the construction of the integral distribution U by first choosing an arbitrary vector field U1 and
then aiming for another vector field U2 such that [U1, U2] ∈ Tι(V[Rq]). During the construction
of the field U2 we regard the components of the vector ζ1 = ζ̂1 as given parameters and the
components of ζ̂2 as the only unknowns of the system
Ξ̂1ζ̂2 = Ξ̂2ζ̂1 −Θ12. (37)
Since the components of ζ̂1 are not considered unknowns, the system (37) must not lead to any
restrictions for the coefficients ζ̂k
1 . Obviously, this is the case if, and only if,
rank Ξ̂1 = rank (Ξ̂1 Ξ̂2). (38)
Assuming that (38) holds, the system (37) is solvable if, and only if, it satisfies the augmented
rank condition
rank Ξ̂1 = rank (Ξ̂1 Ξ̂2 −Θ12). (39)
When we have succeeded in constructing U2, the next step is to seek a further vector field U3
such that [U1, U3] ∈ Tι(V[Rq]) and [U2, U3] ∈ Tι(V[Rq]). Now the components of both vectors
ζ̂1 and ζ̂2 are regarded as given, and the components of ζ̂3 are regarded as the unknowns of the
system
Ξ̂1ζ̂3 = Ξ̂3ζ̂1 −Θ13, Ξ̂2ζ̂3 = Ξ̂3ζ̂2 −Θ23.
Now this system is not to restrict the components of both ζ̂1 and ζ̂2 any further. The interrela-
tions between the ζ̂i following from the condition (35) for the existence of integral distributions,
given in Remark 5, are taken care of by contracting Ξ3 into Ξ̂3. This implies that now the rank
condition
rank
(
Ξ̂1
Ξ̂2
)
= rank
(
Ξ̂1 Ξ̂3 0
Ξ̂2 0 Ξ̂3
)
has to be satisfied. If it is, then for 1 ≤ c ≤ C = dimV ′[Rq]− dimV[Rq] the system
Θc
13 − Ξ̂c
3kζ
k
1 + Ξ̂c
1kζ
k
3 = 0, Θc
23 − Ξ̂c
3kζ
k
2 + Ξ̂c
2kζ
k
3 = 0
is solvable if, and only if, the augmented rank condition
rank
(
Ξ̂1
Ξ̂2
)
= rank
(
Ξ̂1 Ξ̂3 0 −Θ13
Ξ̂2 0 Ξ̂3 −Θ23
)
holds. Now we proceed by iteration. Given j − 1 vector fields U1, U2, . . . , Uj−1 of the required
form spanning an involutive subdistribution of Tι(V[R1]), we construct the next vector field Uj
by solving the system
Ξ̂1ζ̂j = Ξ̂j ζ̂1 −Θ1j ,
· · · · · · · · · · · · · · · · · · · · · (40)
Ξ̂j−1ζ̂j = Ξ̂j ζ̂j−1 −Θj−1,j .
Again we consider only the components of the vector ζ̂j unknowns, and the system (40) must
not imply any further restrictions on the components of the vectors ζ̂i for 1 ≤ i < j. The
corresponding rank condition is
rank
Ξ̂1
Ξ̂2
...
Ξ̂j−1
= rank
Ξ̂1 Ξ̂j
Ξ̂2 Ξ̂j 0
... 0
. . .
Ξ̂j−1 Ξ̂j
. (41)
Existence and Construction of Vessiot Connections 23
Assuming that it holds, the equations (40) are solvable and yield solutions for the components
of ζ̂j if, and only if, the system satisfies the augmented rank condition
rank
Ξ̂1
Ξ̂2
...
Ξ̂j−1
= rank
Ξ̂1 Ξ̂j −Θ1j
Ξ̂2 Ξ̂j 0 −Θ2j
... 0
. . .
...
Ξ̂j−1 Ξ̂j −Θj−1,j
. (42)
Remark 6. We can simplify our calculations for a differential equation Rq, represented by
uα
µ = φα
µ(x,u, û(q)) where each equation is solved for a principal derivative (all pairwise different)
uα
µ with |µ| = q, and where û(q) denotes the set of the remaining, thus parametric, derivatives
of order q and less. (Example 1 discusses a system of that kind.) Now we can use the local
coordinates on Rq. For the generators of the symbol Nq we may choose for (α, µ) 6∈ B the vector
fields Ȳ α
µ = ∂uα
µ
; as a basis for the complement H ⊆ V[Rq], we can take the vector fields
X̄i = ∂
xi +
m∑
α=1
∑
0≤|µ|<q
uα
µ+1i
∂uα
µ
,
which satisfy equation (27). For the vector fields Wa which appear in equation (25) we may
choose the contact vector fields Cµ
α where (α, µ) ∈ B. The further procedure, solving first the
structure equations, which now take the simple form (26), to find the generators of V ′(Rq) and
then solving equation (33) for the coefficient functions ζj , is the same as in the case where the
representation is not in the above solved form.
Example 7. Before we proceed with our theoretical analysis, let us demonstrate with a concrete
differential equation that in general we cannot expect that the above outlined step-by-step
construction of integral distributions works. In order to keep the size of the example reasonably
small, we use the second-order equation1 R2 defined by the system uxx = αu and uyy = βu
with two real constants α, β. Its symbol N2 is not involutive. However, one easily proves that
R2 is formally integrable for arbitrary choices of the constants α, β. One readily computes that
the Vessiot distribution V[R2] is generated by the following three vector fields on R2:
X1 = ∂x + ux∂u + αu∂ux + uxy∂uy , X2 = ∂y + uy∂u + uxy∂ux + βu∂uy , Y1 = ∂uxy .
They yield as structure equations for V[R2]:
[X1, X2] = βux∂uy − αuy∂ux , [X1, Y1] = −∂uy , [X2, Y1] = −∂ux .
For the construction of a two-dimensional integral distribution U ⊂ V[R2] we make as above
the ansatz Ui = Xi + ζ1
i Y1 with two coefficients ζ1
i . As we want to perform a step-by-step
construction, we assume that we have chosen some fixed value for ζ1
1 and try now to determine ζ1
2
such that [U1, U2] ≡ 0 mod V[R2]. Evaluation of the Lie bracket yields the equation
[U1, U2] ≡ (βux − ζ1
2 )∂uy − (αuy − ζ1
1 )∂ux mod V[R2]. (43)
A necessary condition for the vanishing of the right side is that ζ1
1 = αuy. Hence one cannot
choose this coefficient arbitrarily, as we assumed, but (43) determines both functions ζ1
i uniquely.
Note that the conditions on the coefficients ζ1
i imposed by (43) are trivially solvable and thus R2
possesses an integral distribution (which is even involutive). The problem is that it could not
be constructed systematically with the above outlined step-by-step process.
1It could always be rewritten as a first-order one satisfying the assumptions made above, and the phenomenon
we want to discuss is independent of this transformation.
24 D. Fesser and W.M. Seiler
The following theorem links the satisfaction of the rank conditions (41) and (42), and thus
the solvability of the algebraic system (33) by the above described step-by-step process, with
intrinsic properties of the differential equation Rq and its symbol Nq. It represents an existence
theorem for integral distributions.
Theorem 2. Assume that δ-regular coordinates have been chosen for the local representation of
the differential equation Rq. Then the rank condition (41) is satisfied for all 1 ≤ j ≤ n if, and
only if, the symbol Nq is involutive. The augmented rank condition (42) holds for all 1 ≤ j ≤ n
if, and only if, the differential equation Rq is involutive.
The proof of Theorem 2 is given in Appendix A, as parts of it are rather technical. First, we
transform the differential equation into an equivalent first-order system with a representation in
the reduced Cartan normal form (10) and the same Cartan characters. This is always possible;
see, for example, [23, Proposition A.3.1]. The basic idea of the proof is then that the rank
condition (41) is equivalent to the vanishing of the obstructions to involution of the symbol in
Lemma 2, and the augmented rank condition (42) is equivalent to the vanishing of the remaining
integrability conditions (recall that Lemma 2 gives all obstructions to involution only if the used
coordinates are δ-regular so that this assumption is necessary in Theorem 2). The concrete
implementation of this idea requires some technical considerations concerning the transformation
of the matrices (41) and (42) into row echelon form, working out their contractions and analysing
the interrelation between these operations.
Example 8. We demonstrate the role of δ-regularity of the coordinates with the wave equation
uxy = 0 in characteristic coordinates which are not δ-regular. A straightforward computation
yields as generators for the Vessiot distribution the fields X1 = C
(2)
1 , X2 = C
(2)
2 , Y1 = ∂uxx
and Y2 = ∂uyy . Following our approach, we make the ansatz U1 = X1 + ζ1
1Y1 + ζ2
1Y2 and
U2 = X2 +ζ1
2Y1 +ζ2
2Y2. Evaluation of the Lie bracket [U1, U2] yields now the algebraic equations
ζ2
1 = 0 and ζ1
2 = 0 and the differential equations U1(ζ1
2 )− U2(ζ1
1 ) = 0 and U1(ζ2
2 )− U2(ζ2
1 ) = 0.
As in Example 7, we see that the step-by-step process is broken, since we obtain a condition on
the coefficient ζ2
1 = 0 which we should be able to consider a parameter here.
At this point, we have proven that integral distributions within the Vessiot distribution
exist if, and only if, the algebraic conditions (33) are solvable, and that this is equivalent to the
augmented rank condition (41) being satisfied. This in turn is the case precisely if the differential
equation is involutive. Thus we have characterised the existence of Vessiot connections for
Vessiot’s [28] step-by-step approach.
8 The existence theorem for flat Vessiot connections
There remains to analyse the solvability, if we add the differential system (34). Its solvability
is equivalent to the existence of flat Vessiot connections in that each flat Vessiot connection
of R1 corresponds to a solution of the combined system (33), (34). We first note that the set of
differential conditions (34) alone is again an involutive system.
Proposition 7. The differential conditions (34) alone represent an involutive differential equa-
tion of first order.
The proof, which is somewhat technical, is given in Appendix B. It is based on an evaluation
of the Jacobi identity for the Lie brackets of the fields Ui.
If the original equation R1 is analytic, then the quasi-linear system (34) is analytic, too. Thus
we may apply the Cartan–Kähler theorem (Theorem 1) to it which guarantees the existence of
solutions. The problem is that the combined system (33), (34) is not necessarily involutive, as
Existence and Construction of Vessiot Connections 25
the prolongation of the algebraic equations (33) may lead to additional differential equations.
Instead of analysing the effect of these integrability conditions, we proceed as follows. If we as-
sume that R1 is involutive, then we know from Theorem 2 that the algebraic equations (33) are
solvable. Actually, the proof of the Theorem 2 produces an explicit row echelon form of the sys-
tem matrix given on the right hand side of equation (41). Now we use the interrelations between
the unknowns ζ
(α,h)
` , where (α, h) 6∈ B, to eliminate in the differential conditions (34) some of
them by expressing them as linear combinations of the remaining ones. Let 1 ≤ i < j ≤ n; then
for all (α, i) where β
(i)
1 + 1 ≤ α ≤ β
(j)
1 we find
ζ
(α,i)
j =
j∑
k=1
m∑
γ=β
(k)
1 +1
Ck
γ (φα
j )ζ(γ,k)
i + C
(1)
i (φα
j ), (44)
while for all (α, i) where β
(j)
1 + 1 ≤ α ≤ m the correspondence (35), given in Remark 5, holds
again (see [6, Corollary 3.3.23] for the detailed calculation). In this way, we plug the algebraic
conditions into the differential conditions and thus get rid of them.
Example 9. For Example 1, where i = 1 and j = 2, we deduce for
U1 = X1 + ζvx
1 ∂vx + ζwx
1 ∂wx and U2 = X2 + ζvx
2 ∂vx + ζwx
2 ∂wx
from the algebraic condition (37) that −ζvx
2 = −ζvx
1 and −ζwx
2 = −ζwx
1 . This is equation (44) for
(α, i) ≡ vx and (α, i) ≡ wx. The rank conditions (38) and (39) are satisfied because vx−wt = 0,
so that the matrix (Ξ1 Ξ2 − Θ12) has a vanishing first row. (There is no contraction of
matrices for j = 2.) There are no interrelations like those in equation (35) because there is
no index value α for which both ι∗(∂uα
1
) and ι∗(∂uα
2
) would be symbol fields. The differential
conditions (34) are
U1(ζvx
2 )− U2(ζvx
1 ) = 0, U1(ζwx
2 )− U2(ζwx
1 ) = 0,
and by substituting ζvx
2 = ζvx
1 and ζwx
2 = ζwx
1 , we can drop the algebraic conditions from the
system. Since n = 2, there is no further procedure.
We can now prove the following existence theorem for flat Vessiot connections.
Theorem 3. Assume that δ-regular coordinates have been chosen for the local representation of
the analytic differential equation R1. Then the combined system (33), (34) is solvable.
The idea of the proof is this: following the strategy we have just outlined, we eliminate some
of the unknowns ζ̂k
i . Because of the simple structure of (34), it turns out that we must take
a closer look only at those equations where the leading derivative is of one of the unknowns we
eliminate. A somewhat lengthy but straightforward computation shows that these equations
actually vanish. The remaining equations still form an involutive system. Thus we eventually
arrive at an analytic involutive differential equation for the coefficient functions ζ̂k
i which is
solvable according to the Cartan–Kähler theorem. The details are worked out in Appendix C.
Example 10. Consider the first-order equation
R1 :
{
ut = vt = wt = us = 0, vs = 2ux + 4uy,
ws = −ux − 3uy, uz = vx + 2wx + 3vy + 4wy
in the five independent variables x, y, z, s, t and the three dependent variables u, v, w. It is
formally integrable, and its symbol is involutive with dimN1 = 8. Thus R1 is an involutive
equation. For the matrices Ξi, all of which are 3× 8-matrices, we find
Ξ1 =
(−1 0 0 0 0 0 0 0
0 −1 0 0 0 0 0 0
0 0 −1 0 0 0 0 0
)
, Ξ2 =
(
0 0 0 −1 0 0 0 0
0 0 0 0 −1 0 0 0
0 0 0 0 0 −1 0 0
)
,
26 D. Fesser and W.M. Seiler
Ξ3 =
( 0 −1 −2 0 −3 −4 0 0
0 0 0 0 0 0 −1 0
0 0 0 0 0 0 0 −1
)
, Ξ4 =
(
0 0 0 0 0 0 0 0
−2 0 0 −4 0 0 0 0
1 0 0 3 0 0 0 0
)
, Ξ5 = 03×8.
For the first two steps in the construction of the fields Ui, the rank conditions are trivially
satisfied even for the non-contracted matrices. But not so in the third step where we have in the
row echelon form of the arising 9×32-matrix in the 7th row zero entries throughout except in the
12th column (where we have −2) and in the 17th column (where we have 2). As a consequence,
we obtain the equality ζ4
1 = ζ1
2 and the rank condition for this step does not hold. However, since
both ux and uy are parametric derivatives and in our ordering Y1 = ι∗(∂ux) and Y4 = ι∗(∂uy),
this equality is already taken into account in our reduced ansatz and for the matrices Ξ̂i the
rank condition is satisfied.
Note that the rank condition is first violated when the rank reaches the symbol dimension
(which is 8). From then on, the rank of the left matrix in (41) stagnates at dimN1 while the
rank of the augmented matrix rises further. The entries breaking the rank condition differ by
their sign, while their corresponding coefficients in Lemma 2 are collected into one sum and thus
vanish. Here it shows that contracting the matrices, as explained in Remark 5, is necessary.
9 Conclusions
Vessiot’s [28] original motivation for the introduction of his theory was to provide an alternative
proof of the Cartan–Kähler theorem (the same holds for Stormark’s presentation [24] of the
theory). If one takes this point of view, then one may say that our proof of Theorem 3 is
a “cheat”, as it uses the Cartan–Kähler theorem instead of proving it. However, in our opinion,
this point of view is the main reason why the corresponding proofs in [24, 28] are so difficult to
read (and actually incomplete as they neglect that the involutivity of the system is a necessary
condition). The direct proof of the Cartan–Kähler theorem (for differential equations) given in
textbooks on the formal theory like [19, 23] is much simpler and more transparent; in particular,
it makes clear where involution (as opposed to mere formal integrability) is needed. Thus, if the
only goal consisted of proving such an existence and uniqueness theorem, then there would be
no need to bother with Vessiot’s theory.
We believe that the real value of Vessiot’s theory lies in the fact that it provides via the
distribution V[Rq] an additional geometric structure on an involutive differential equation Rq
which is very useful for the further analysis of the equation, that is, after its solvability has been
established. One possible application is the investigation of certain forms of singular behaviour
of solutions. In fact, the classical works of Arnold and collaborators (see [2] for an elementary
introduction) on implicit ordinary differential equations are based on the Vessiot distribution
(without using this terminology); for some further works also in the context of partial differential
equations or numerical analysis see, for example, [16, 25, 26]. The basic idea here is that such
behaviour mainly stems from the fact that at some points the considered involutive distributions
cease to be transversal to the fibration Rq → X so that integral manifolds can no longer be
interpreted as prolonged solutions in the classical sense. However, it often makes sense to
consider them a generalised form of (potentially multi-valued) solutions.
Another application concerns symmetry theory. In classical symmetry theory one always
considers the differential equationRq as a submanifold of Jqπ and looks then for diffeomorphisms
of Jqπ which are (i) compatible with the contact structure of Jqπ and (ii) leave Rq invariant.
It was only fairly late realized [1] that this approach yields only what is now called external
symmetries and that in general further useful symmetries may exist. Using Vessiot’s theory, we
may represent a differential equation as the pair (Rq,V[Rq]), that is, as a manifold together with
a distribution on it – without any recourse to an ambient space (all relevant properties of Jqπ are
captured in the Vessiot distribution V[Rq]). This point of view yields automatically all internal
symmetries (and is equivalent to the symmetry theory of exterior differential systems).
Existence and Construction of Vessiot Connections 27
We mentioned already in the Introduction that Vessiot’s theory takes an intermediate po-
sition between the formal theory and the theory of exterior differential systems. It allows for
the transfer of ideas from the latter to the former one without the need to rewrite a differential
equation as an exterior system. For example, [23, Section 9.5] contains a formulation of prolon-
gation structures without exterior forms using the Vessiot distribution instead2. It should also be
possible to give an explicit proof of the equivalence of the formal theory and the Cartan–Kähler
theory on the basis of the results presented here.
Finally, we comment on the practical side of the Vessiot theory. Throughout this article, we
made a number of assumptions on the treated differential equation: we assumed that it is first
rewritten as a first-order system, then that all present algebraic equations are explicitly solved
and finally that each equation is solved for its principal derivative. While, from a theoretical
point of view, these operations are always possible under fairly mild regularity assumptions,
usually they cannot be performed effectively. However, the made assumptions are only needed
in order to be able to prove results like Lemma 2 on the explicit form of the obstructions to
involution. These expressions are already bad enough with our simplifying assumptions; for
more general systems they would hardly be manageable.
A concrete differential equation need not be transformed to such a special normal form in
order to apply the Vessiot theory. Even for fully implicit equations of arbitrary order, the
determination of all integral distributions requires only linear algebra and is easily implemented
in a computer algebra system. In fact, already Fackerell [5] reported of an implementation of
Vessiot’s theory. Within a recent project thesis [8], Globke implemented large parts of the theory
in MuPAD. The problems and limitations are here exactly the same as in implementations of
the Cartan–Kähler theory (see, for example, [10]).
A Proof of Theorem 2
In this section of the appendix, we prove Theorem 2, the existence theorem for integral distribu-
tions. It is in principle by straightforward matrix calculation and involves a tedious distinction
of several cases and subcases: we have to compare the entries in the matrix on the right hand
side of equation (42) for step j after turning it into row echelon form with the integrability
conditions and the obstructions to involution as they are given in Lemma 2. For this purpose,
we fix 1 ≤ i < j and consider the block Ξ̂i and the entries to its right in the complete matrix.
Since for a differential equation with an involutive symbol the obstructions to involution va-
nish and for an involutive differential equation the integrability conditions vanish, too, it follows
that the augmented rank condition, stated in equation (42), is equivalent to the differential
equation being involutive, which in turn is the case if, and only if, the algebraic conditions (33)
are satisfied, which is necessary and sufficient for the existence of integral distributions within
the Vessiot distribution.
If the order of the differential equation is q > 1, transform it into an equivalent first-order
equation; for the details of this procedure, see [6, Subsection 2.5.1] or [23]. If Rq is involutive,
then so is R1 (see [23] for a straightforward proof of this). We assume this first-order equation is
represented in reduced Cartan normal form (10). We first prove the rank condition (41) for the
homogeneous system. (The proof for the augmented rank condition (42) follows.) We proceed
in two steps: We consider the complete matrix at the jth step,
Ξ1 Ξj
Ξ2 Ξj 0
... 0
. . .
Ξj−1 Ξj
. (45)
2Both symmetry theory and prolongation structures were also discussed by Fackerell [5] via Vessiot theory.
28 D. Fesser and W.M. Seiler
(It differs from the matrix on the right hand side of equation (41) in that it is not yet contracted.)
The complete matrix at the jth step is built from (j−1)j blocks: the stack of j−1 matrices Ξi,
1 ≤ i ≤ j − 1, on the left, with each of the Ξi having another j − 1 blocks to its right, the ith
of which being Ξj and all the others being zero.
For easier reference, let, for 1 ≤ i, k ≤ j − 1, be [i, k] the kth block right of a Ξi. Then, for
all 1 ≤ i, k ≤ j − 1, we have
[i, k] =
{
Ξj , i = k,
0m×r, i 6= k.
For convenience, we set [i, 0] := Ξi. Let, for 1 ≤ g ≤ h ≤ n,
[Ξi]
g...h (46)
denote the matrix that results from writing the block matrices [Ξi]
g, [Ξi]
g+1, . . . , [Ξi]
h (from
left to right) next to each other. Let [i, k]h be the hth block of columns in [i, k], and let, for
1 ≤ g ≤ h ≤ n, in analogy to the shorthand (46),
[i, k]g...h
denote the matrix that results from writing the matrices [i, k]g, [i, k]g+1, . . . , [i, k]h (from left
to right) next to each other.
If M is any a× b-matrix, then let d
c [M ] be the matrix made from the rows with indices (that
is, labels) c to d, [M ]fe the matrix made from the columns with indices e to f and d
c [M ]fe the
matrix made from the entries in the rows with indices c to d and in the columns with indices
from e to f . If the columns of M are grouped into blocks and g denotes which blocks are meant,
then we write c
1[M ]g (with g up right) to show that the first c upper rows are being selected,
and we write b
d[M ]g (with g below right) to show that the last b − d + 1 lower rows are being
selected. For the block matrix made from M by selecting the rows indexed c to d within the
block of columns labelled g, we write d
c [M ]g. This notation is redundant in that the position
of g does not give new information, but in the calculations to come it increases readability.
We first turn the complete matrix into row echelon form. Then we contract columns and
consider the effect.
The matrices Ξi are shown in equation (31) and have a block-structure of the form(
aαr bαs cαt
0S×R −1S dst
)
=: (gij : 1 ≤ i ≤ A + S, 1 ≤ j ≤ R + S + T ),
where A, R, S, T are adequate natural numbers or zero, and the blocks are(
aαr : 1 ≤ α ≤ A, 1 ≤ r ≤ R
)
,
(
bαs : 1 ≤ α ≤ A, 1 ≤ s ≤ S
)
,(
cαt : 1 ≤ α ≤ A, 1 ≤ t ≤ T
)
,
(
dst : 1 ≤ s ≤ S, 1 ≤ t ≤ T
)
,
the S × S unit matrix 1S or the zero matrix 0S×R. (Here the index r is just some index and
not supposed to be the dimension of any symbol.) Then the substitution(
aαr bαs cαt
)
←
(
aαr bαs cαt
)
+ (bαs) ·
(
0S×R −1S dst
)
(47)
transforms the matrix (gij) into(
aαr 0A×S cαt +
∑S
s=1 bαsdst
0S×R −1S dst
)
.
Existence and Construction of Vessiot Connections 29
For the transformation of the complete matrix into row echelon form, we use this obvious method.
Let 1 < j ≤ n be given. Choose 1 ≤ i < j. Now we eliminate all the non-trivial entries in the
rows
β
(i)
1
1 [Ξi [i, 1] [i, 2] . . . [i, j − 1]] .
We turn to the block β
(i)
1
1 [Ξi] first. Non-trivial entries therein may appear according to equa-
tion (31) only for 1 ≤ h ≤ i in the blocks β
(i)
1
1 [Ξi]
h. So we fix 1 ≤ h ≤ i and consider the
block β
(i)
1
1 [Ξi]
h. Now set
(aαr) = β
(i)
1
1 [Ξi]
1...h−1,
(bαs) = β
(i)
1
1 [Ξi]
h,
(cαt) =
β
(i)
1
1 [[Ξi]h+1...n[i, 1][i, 2] . . . [i, j − 1]], (48a)
(dst) = m
β
(h)
1 +1
[[Ξh]h+1...n[h, 1][h, 2] . . . [h, j − 1]], (48b)
−1S = −1
α
(h)
1
,
0S×R = 0
α
(h)
1 ×
∑h−1
l=1 α
(l)
1
.
Then it follows that the substitution (47) leaves (aαr) unchanged, turns (bas) into zero as required
and makes (cαt) into (cαt +
∑S
s=1 bαsdst). According to (48b) the entries
∑S
s=1 bαsdst here
are of two types: from m
β
(h)
1 +1
[Ξh]h+1...n, the left part of (dst), we have the entries in (bαs) ·
m
β
(h)
1 +1
[Ξh]h+1...n, and from m
β
(h)
1 +1
[[h, 1][h, 2] . . . [h, j − 1]], the right part of (dst), we have those
of (bαs) · mβ(h)
1 +1
[h, l] where 1 ≤ l ≤ j − 1.
Consider the cαt +
∑S
s=1 bαsdst where the factors dst are from the left part of (dst): according
to equation (31), m
β
(h)
1 +1
[Ξh]h+1...n = 0, so
(bαs) · mβ(h)
1 +1
[Ξh]h+1...n = 0.
It follows cαt+
∑S
s=1 bαsdst = cαt, so that all entries in β
(i)
1
1 [Ξi]
h+1...n, the left part of (cst), remain
unchanged through the elimination of (bαs) = β
(i)
1
1 [Ξi]
h.
For the rest of the proof, we consider the remaining entries cαt +
∑S
s=1 bαsdst ,where the
factors dst are from m
β
(h)
1 +1
[[h, 1][h, 2] . . . [h, j − 1]], the right part of (dst). Fix 1 ≤ l ≤ j − 1 and
consider (bαs) · mβ(h)
1 +1
[h, l].
There are two cases: h = l and h 6= l. For h 6= l we have [h, l] = 0 according to the definition
of the complete matrix (45), thus m
β
(h)
1 +1
[h, l] = 0 and so
(bαs) · mβ(h)
1 +1
[h, l] = 0.
Again it follows cαt +
∑S
s=1 bαsdst = cαt, so that for h 6= l all entries in
β
(i)
1
1 [[i, 1][i, 2] . . . [i, j − 1]],
the right part of (cαt), remain unchanged through the elimination of (bαs) = β
(i)
1
1 [Ξi]
h, too.
30 D. Fesser and W.M. Seiler
For h = l, we have [h, h] = Ξj . The structure of Ξj , which is (31) with i replaced by j,
implies that non-vanishing entries are possible in the blocks m
β
(h)
1 +1
[h, h]k where 1 ≤ k ≤ j. In
fact
m
β
(j)
1 +1
[h, h]k = 0 for 1 ≤ k ≤ j − 1 and (49a)
m
β
(j)
1 +1
[h, h]j = −1
α
(j)
1
. (49b)
The entries
∑S
s=1 bαsdst for those dst within m
β
(h)
1 +1
[h, h]k are
(bαs) · mβ(h)
1 +1
[h, h]k = β
(i)
1
1 [Ξi]
h · m
β
(h)
1 +1
[Ξj ]k =
(
S∑
s=1
Ch
β
(h)
1 +s
(φα
i )Ck
β
(k)
1 +t
(φβ
(h)
1 +s
j )
)
. (50)
This matrix has A = β
(i)
1 rows and T = α
(k)
1 columns. We consider its entry in row α and
column t. Setting γ := β
(h)
1 + s, δ := β
(k)
1 + t and using S = α
(h)
1 in (50), this entry is
β
(h)
1 +α
(h)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ck
δ (φγ
j ) (51)
for all 1 ≤ k ≤ j. Some of the β
(h)
1 + α
(h)
1 = m summands vanish because of the special
entries (49). As a consequence, from β
(j)
1 + 1 on, of all the summands Ch
γ (φα
i )Ck
δ (φγ
j ) in (51) at
most one remains: none for k 6= j, exactly one for k = j, namely the one for γ = β
(j)
1 + t = δ.
Using the Kronecker-delta, for all β
(j)
1 + 1 ≤ δ ≤ m we have
Ch
δ (φα
i )Cj
δ (φ
δ
j) = δkj · Ch
δ (φα
i ).
Now (51) becomes for all 1 ≤ α ≤ β
(i)
1 and all β
(k)
1 + 1 ≤ δ ≤ m
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ck
δ (φγ
j ) + δkj · Ch
δ (φα
i ). (51′)
These are the terms
∑S
s=1 bαsdst in the case h = l when 1 ≤ k ≤ j. Now for the subcase
j + 1 ≤ k ≤ n: these blocks m
β
(h)
1 +1
[h, h]k = m
β
(h)
1 +1
[Ξj ]k, again on account of the structure of Ξj ,
which is (31) with i replaced by j, are zero (if they exist at all; they do not for k with α
(k)
1 = 0).
So in this case we have
(bαs) · mβ(h)
1 +1
[Ξj ]j+1...n = 0,
which contains the terms
∑S
s=1 bαsdst = 0. As a consequence, here, in the case h = l, the
terms cαt +
∑S
s=1 bαsdst are of the following form. The cαt of interest in line (48a) are those
in the block β
(i)
1
1 [i, h]. For h 6= i, we have [i, h] = 0, thus β
(i)
1
1 [i, h] = 0. Non-trivial entries cαt
are possible only for h = i; in that case [i, i] = Ξj , thus β
(i)
1
1 [i, h] = β
(i)
1
1 [Ξj ]. According to the
structure of Ξj , which is (31) with i replaced by j, the only entries cαt within the block β
(i)
1
1 [Ξj ]
which may not vanish are, for 1 ≤ k ≤ j, those of the form −Ck
δ (φα
j ). They make up the block
β
(i)
1
1 [Ξj ]
1...j = β
(i)
1
1 [i, i]1...j .
Existence and Construction of Vessiot Connections 31
k
1 2 . . . i− 1 i i + 1 . . . j − 1 j
1 3.(d2)
2 3.(d2) 3.(d1) 3.(c) 3.(b) 3.(a)
h
...
. . .
i− 1 3.(d2)
i 2.(a) 2.(b)
Figure 2. Possible combinations of h and k in the terms uδ
hk used as labels for columns in the contracted
matrix. Each term uδ
hk corresponds to one subcase in the consideration of cases 2. h = i and 3. h < i.
Each of these subcases refers to one of the terms in square brackets in Lemma 2 as follows: 2.(a) –
line (17), 2.(b) – line (19), 3.(a) – line (18), 3.(b) – line (16), 3.(c) – line (15), 3.(d1) – line (14), 3.(d2) –
line (13).
So for any row index α and any column index δ = β
(k)
1 + t we have cαt = −δhiC
k
δ (φα
j ) as the
most general form of an entry.
Since (51′) is
∑S
s=1 bαsdst, it follows that cαt +
∑S
s=1 bαsdst is
−δhiC
k
δ (φα
j ) +
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ck
δ (φγ
j ) + δkjC
h
δ (φα
i ). (52)
Since 1 ≤ α ≤ β
(i)
1 and β
(k)
1 + 1 ≤ δ ≤ m, any term in the row echelon form of the complete
matrix without a −1 in a negative unit block somewhere to its left has this form or is a zero
in β
(i)
1
1 [Ξi]
i+1...n for some i < j (in which case it remains zero throughout the elementary row
transformations and so does not influence the rank). This means, if all these expressions vanish
(when contracted), the rank condition is satisfied. To show that they do vanish (when con-
tracted) for a system with an involutive symbol, we consider them as the new entries in β
(i)
1
1 [i, h]
with 1 ≤ h ≤ j. Now with regard to the relation between h and i there are three cases: h > i,
h = i and h < i. We consider them in that order.
1. Let h > i. Then according to the structure of Ξi, (31), all the Ch
γ (φα
i ) and Ch
δ (φα
i ) in (52)
vanish. Since h 6= i, δhi = 0, thus all of (52) vanishes.
2. Let h = i. We shall consider several subcases for h = i, and for h < i after that, which
may be labelled by second-order derivatives uδ
hk where h and k are combined as shown in
Fig. 2. For fixed δ, any uδ
hk belongs, according to its indices h and k, to exactly one of the
blocks marked by the case denominations 2.(a), 2.(b) and 3.(a) to 3.(d2).
It turns out that (52) is the common form of all the sums that appear in the squared
brackets of Lemma 2 as the coefficients of second-order derivatives uδ
hk. Not only can the
case distinctions of the following argument, 2.(a), 2.(b) and 3.(a) to 3.(d2), be labelled by
these uδ
hk, but in fact a case labelled uδ
hk is dealt with by using the fact that according
to Lemma 2 the coefficient in square brackets of that same uδ
hk vanishes for an involutive
system. The correspondence of the cases and the terms uδ
hk is given in the caption of
Fig. 2.
The cases h = i with subcase k < i and h < i with subcase k < i need not be considered
because the columns of the complete matrix are to be contracted when proving the rank
condition. This contraction concerns those columns of the complete matrix which are
labelled by the same second-order derivative, and each second-order derivative is used
exactly once in the argument.
32 D. Fesser and W.M. Seiler
(a) For the first subcase of h = i let i ≤ k < j. Then (52) becomes
−Ck
δ (φα
j ) +
β
(j)
1∑
γ=β
(i)
1 +1
Ci
γ(φα
i )Ck
δ (φγ
j )
where β
(k)
1 + 1 ≤ δ ≤ m. This vanishes for a system with an involutive symbol according
to Lemma 2, line (17).
(b) For the second subcase of h = i choose k = j. Then (52) becomes
−Cj
δ (φ
α
j ) +
β
(j)
1∑
γ=β
(i)
1 +1
Ci
γ(φα
i )Cj
δ (φ
γ
j ) + Ci
δ(φ
α
i )
where β
(j)
1 + 1 ≤ δ ≤ m. According to Lemma 2, line (19), this vanishes for a system with
an involutive symbol.
3. Let h < i. We consider several subcases with regard to the relation between h < i, j and k.
(a) First choose k = j. Then (52) becomes
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Cj
δ (φ
γ
j ) + Ch
δ (φα
i )
where β
(j)
1 + 1 ≤ δ ≤ m. According to Lemma 2, line (18), this vanishes for a system with
an involutive symbol.
(b) For the second subcase choose i < k < j. Then (52) becomes
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ck
δ (φγ
j )
where β
(k)
1 + 1 ≤ δ ≤ m. According to Lemma 2, line (16), this vanishes for a system with
an involutive symbol.
(c) For the third subcase choose k = i. Then (52) becomes
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ci
δ(φ
γ
j ) (53)
where β
(k)
1 + 1 ≤ δ ≤ m. Since we have h < k = i, for any such δ the cross derivative
uδ
ih = uδ
hi labels two columns in the complete matrix: the one with label δ in β
(i)
1
1 [i, h]i and
the one with label δ in β
(i)
1
1 [i, i]h which according to (52) has the new entries
−δkiC
h
δ (φα
j ) +
β
(j)
1∑
γ=β
(k)
1 +1
Ck
γ (φα
i )Ch
δ (φγ
j ) + δhjC
k
δ (φα
i ) (54)
Existence and Construction of Vessiot Connections 33
where β
(h)
1 + 1 ≤ δ ≤ m. In the current subcase, (54) becomes
−Ch
δ (φα
j ) +
β
(j)
1∑
γ=β
(i)
1 +1
Ci
γ(φα
i )Ch
δ (φγ
j ) (55)
where β
(h)
1 +1 ≤ δ ≤ m. Since the columns of both (53) and (55) are labelled by the same
second-order derivatives, we have to contract them, which means adding their new entries.
For all β
(h)
1 + 1 ≤ δ ≤ β
(i)
1 this yields
−Ch
δ (φα
j ) +
β
(j)
1∑
γ=β
(i)
1 +1
Ci
γ(φα
i )Ch
δ (φγ
j ),
which, for a system with an involutive symbol, vanishes according to Lemma 2, line (15a),
and for all β
(i)
1 + 1 ≤ δ ≤ m
−Ch
δ (φα
j ) +
β
(j)
1∑
γ=β
(i)
1 +1
Ci
γ(φα
i )Ch
δ (φγ
j ) +
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ci
δ(φ
γ
j ),
which, for a system with an involutive symbol, again vanishes according to Lemma 2,
line (15b).
(d) For the fourth subcase choose k < i. Under this assumption, we have to distinguish two
further subcases: k = h and k < h. The subcase k > h need not be considered since
uδ
kh = uδ
hk and any second-order derivative is used only once to label a column in the
contracted matrix.
(d1) First consider k < i and k = h. Then still h < i, furthermore k < j, and (52) becomes
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ch
δ (φγ
j )
where β
(h)
1 + 1 ≤ δ ≤ m. According to Lemma 2, line (13), this vanishes for a system with
an involutive symbol.
(d2) At last consider h < k < i. Then k < j, and (52) becomes
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ck
δ (φγ
j ) (56)
where β
(k)
1 + 1 ≤ δ ≤ m. Since we have h < k < i, for any such δ the cross derivative
uδ
kh = uδ
hk labels two columns in the complete matrix: the one with label δ in β
(i)
1
1 [i, h]k
and the one with label δ in β
(i)
1
1 [i, k]h which according to (52) has the new entries
−δkiC
h
δ (φα
j ) +
β
(j)
1∑
γ=β
(k)
1 +1
Ck
γ (φα
i )Ch
δ (φγ
j ) + δhjC
k
δ (φα
i ) (57)
34 D. Fesser and W.M. Seiler
where β
(h)
1 + 1 ≤ δ ≤ m. In the current subcase, (57) becomes
β
(j)
1∑
γ=β
(k)
1 +1
Ck
γ (φα
i )Ch
δ (φγ
j ) (58)
where β
(h)
1 +1 ≤ δ ≤ m. Since the columns of both (56) and (58) are labelled by the same
second-order derivatives, we have to contract them, which means adding their new entries.
This yields
β
(j)
1∑
γ=β
(k)
1 +1
Ck
γ (φα
i )Ch
δ (φγ
j )
for β
(h)
1 + 1 ≤ δ ≤ β
(k)
1 and vanishes according to Lemma 2, line (14a), and
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Ck
δ (φγ
j ) +
β
(j)
1∑
γ=β
(k)
1 +1
Ck
γ (φα
i )Ch
δ (φγ
j )
for β
(k)
1 + 1 ≤ δ ≤ m and vanishes according to Lemma 2, line (14b).
What we have shown is that in the contracted matrix all the entries without some entry −1
in a negative unit block to their left may be eliminated by elementary row transformations if
the equation has an involutive symbol. Thus under this assumption the rank condition (41) is
satisfied.
Now for the augmented rank condition, equation (42). To transform the augmented complete
matrix into row echelon form, we use for each j and i the same procedure as for the transfor-
mation of the non-augmented complete matrix, except that now the matrices (cαt) and (dst)
are augmented by one more column each as follows. Fix 1 < j ≤ n. Let 1 ≤ i < j. Then,
according to the structure of the augmented complete matrix, for its transformation into row
echelon form we have to eliminate for 1 ≤ h ≤ i the entries in the blocks [Ξi]
h, as we did for the
non-augmented complete matrix given in equation (45). We have to consider the effect of these
transformations on the additional entries which make up the rightmost column in the augmented
complete matrix. These are −Θ1
ij , −Θ2
ij , . . . , −Θm
ij , given in equation (28). Of these entries, only
−Θ1
ij , −Θ2
ij , . . . , −Θ
β
(i)
1
ij are affected (since we eliminate the entries which are in rows 1 to β
(i)
1
of the matrices [Ξi]
h). We add them as the rightmost column in the augmented matrix (cαt).
Now fix 1 ≤ h ≤ i. Then augment the matrix (dst), used in the process of eliminating the
entries in [Ξi]
h, by adding the entries −Θ
β
(h)
1 +1
hj , −Θ
β
(h)
1 +2
hj , . . . , −Θm
hj as its rightmost column,
in accordance with the structure of the augmented complete matrix. Then the substitution (47)
yields as the transformed entries cαt +
∑S
s=1 bαsdst the same as for the transformation of the
non-augmented matrix except, of course, for the new last column. Now let denote t the index
of this last column. Fix some row index 1 ≤ α ≤ β
(i)
1 . Then the entry cαt = −Θα
ij transforms as
follows: For all 1 ≤ h ≤ i we have to add to it
S∑
s=1
bαsdst =
m∑
γ=β
(h)
1 +1
bαγdγt.
Existence and Construction of Vessiot Connections 35
Since here the matrix (bαγ) = [Ξi]
h, this is the product of the row with index α in the matrix [Ξi]
h
and the transpose of the vector (−Θ
β
(h)
1 +1
hj ,−Θ
β
(h)
1 +2
hj , . . . ,−Θm
hj). According to the structure
of [Ξi]
h as defined in equation (30), this equals
m∑
γ=β
(h)
1 +1
Ch
γ (φα
i )Θγ
hj .
The entries Θγ
hj can be taken from equation (28). Since β
(h)
1 +1 ≤ γ ≤ m, we have Θγ
hj = C
(1)
h (φα
j )
if β
(h)
1 + 1 ≤ γ ≤ β
(j)
1 and Θγ
hj = 0 if β
(j)
1 + 1 ≤ γ ≤ m. Therefore we get
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )C(1)
h (φα
j )
as the summand for each h to be added to cαt = −Θα
ij . Since 1 ≤ α ≤ β
(i)
1 , we have −Θα
ij =
−C
(1)
i (φα
j ) + C
(1)
j (φα
i ). Therefore the entry cαt transforms into
−C
(1)
i (φα
j ) + C
(1)
j (φα
i ) +
i∑
h=1
β
(j)
1∑
γ=β
(h)
1 +1
Ch
γ (φα
i )C(1)
h (φγ
j ),
which is the integrability condition in line (12) of Lemma 2, except for the sign. The difference
in sign comes from the fact that the augmented complete matrix describes the system of equa-
tions (40), where the entries from the row with index α in the matrix Ξi are on the opposite side
from the entries from the row with index α in the matrix Ξj and the inhomogeneous term −Θα
ij ,
while in Lemma 2 for each i, j and α the corresponding equation is set to zero, if the differential
equation is involutive.
This means that the augmented rank condition holds for all 1 ≤ i < j ≤ n if, and only if, the
equation has an involutive symbol and is formally integrable.
B Proof of Proposition 7
We are going to show that the system (34) is involutive. Consider the basis (Ui : 1 ≤ i ≤ n)
of the distribution U given by Ui = Xi + ζp
i Yp with yet undetermined coefficient functions
ζp
i ∈ F(R1). For all 1 ≤ i ≤ n and 1 ≤ p ≤ r, the independent variables of the functions ζp
i
are the coordinates on R1, which are x,u and all uα
h such that (α, h) 6∈ B. To apply a vector
field Uj = ∂xj + · · · to a function ζp
i includes a derivation with respect to xj . We order the
independent variables such that if j > i, then xj is greater than xi, and each xi is greater than
all the variables uα and uα
h where (α, h) 6∈ B. For any equation Hp
ij within the system (34), the
application of the vector field Uj = ∂xj + · · · to ζp
i yields ∂ζp
i /∂xj as the leader of that equation;
therefore equation Hp
ij is of class j, and the equations of maximal class are Hp
in; the equations
of second highest class in the system are Hp
in−1 and so on. There are only equations Hp
ij of a
class indicated by some index 2 ≤ j ≤ n.
From the Jacobi identity for vector fields Ui, Uj and Uk where 1 ≤ i < j < k ≤ n, we have
[Ui, [Uj , Uk]] + [Uj , [Uk, Ui]] + [Uk, [Ui, Uj ]] = 0.
36 D. Fesser and W.M. Seiler
The structure equations (32) for the vector fields Uh and the definitions of Gc
ij and Hp
ij in
equations (33) and (34) imply that this is
0 = [Ui, G
c
jkZc + Hp
jkYp] + [Uj , G
c
kiZc + Hp
kiYp] + [Uk, G
c
ijZc + Hp
ijYp]
= Gc
jk[Ui, Zc] + Ui(Gc
jk)Zc + Hp
jk[Ui, Yp] + Ui(H
p
jk)Yp
+ Gc
ki[Uj , Zc] + Uj(Gc
ki)Zc + Hp
ki[Uj , Yp] + Uj(H
p
ki)Yp
+ Gc
ij [Uk, Zc] + Uk(Gc
ij)Zc + Hp
ij [Uk, Yp] + Uk(H
p
ij)Yp.
The combined system (33, 34) means that all Gc
ab = 0 and all Hp
ab = 0 which implies that U is
involutive, which it is, being in triangular form, exactly if all [Ua, Ub] = 0. This leaves only
0 = {Ui(Gc
jk) + Uj(Gc
ki) + Uk(Gc
ij)}Zc + {Ui(H
p
jk) + Uj(H
p
ki) + Uk(H
p
ij)}Yp.
As part of a basis for V ′[Rq], the vector fields Zc and Yp are linearly independent, which means
their coefficients must vanish individually. So in particular
Ui(H
p
jk) + Uj(H
p
ki) + Uk(H
p
ij) = 0.
Under the assumption i < j < k, the term Uk(H
p
ij) contains derivations with respect to xk
of Ui(ζ
p
j ) and Uj(ζ
p
i ). Thus, according to our order, this is a non-multiplicative prolongation,
and the remaining terms are multiplicative prolongations. But since any non-multiplicative
prolongation within the system (34) must be of such a form, it is a linear combination of
multiplicative prolongations. Therefore, no integrability conditions arise from cross-derivatives
(and none arise from a prolongation of lower order equations since all equations of the system
are of first order).
If we set ∂ζp
i /∂xj =: (ζp
i )j for the leaders and
Ũj(ζ
p
i ) := Uj(ζ
p
i )− (ζp
i )j
and solve each equation of the system (34) for its leader, then it takes the form
(ζp
1 )n = U1(ζp
n)− Ũn(ζp
1 ),
(ζp
2 )n = U2(ζp
n)− Ũn(ζp
2 ),
· · · · · · · · · · · · · · · · · · · · · · · ·
(ζp
n−1)n = Un−1(ζp
n)− Ũn(ζp
n−1),
(ζp
1 )n−1 = U1(ζ
p
n−1)− Ũn−1(ζ
p
1 ),
(ζp
2 )n−1 = U2(ζ
p
n−1)− Ũn−1(ζ
p
2 ), (34∗)
· · · · · · · · · · · · · · · · · · · · · · · ·
(ζp
n−2)n−1 = Un−2(ζ
p
n−1)− Ũn−1(ζ
p
n−2),
· · · · · · · · · · · · · · · · · · · · · · · ·
(ζp
2 )3 = U2(ζ
p
3 )− Ũ3(ζ
p
2 ),
(ζp
1 )3 = U1(ζ
p
3 )− Ũ3(ζ
p
1 ),
(ζp
1 )2 = U1(ζ
p
2 )− Ũ2(ζ
p
1 );
here for each line 1 ≤ p ≤ r. Therefore the system (34∗) is in Cartan normal form given in
Definition 7.
Now one can prove (see [6, Lemma 2.4.29]) that such a differential equation R1 is involutive
if, and only if, all non-multiplicative prolongations of the equations (8a)–(8c) and all formal
derivatives with respect to all the xi of the algebraic equations (8d) are dependent on the
equations of the system (8) and its multiplicative prolongations only. Therefore, it follows that
the system (34) is involutive.
Existence and Construction of Vessiot Connections 37
C Proof of Theorem 3
We consider the combined system of algebraic and differential conditions (33), (34) and want
to show that it has a solution. We follow the strategy outlined above and eliminate some of
the unknowns ζp
` . As we consider each of the equations of (34) as being solved for its derivative
∂ζ
(β,h)
i /∂xj of highest class j, as given in equation (34∗), we must take a closer look only at those
equations where this leading derivative is of one of the unknowns we eliminate. The structure of
the vectors ζi, given in equations (35) and (44), shows which ones these are. Let k be such that
2 ≤ k ≤ n. Then for the subsystem of the equations of class k in the system (34), the equations
which hold the following terms are concerned:
Uk
(
ζ
(β,1)
2
)
,
Uk
(
ζ
(β,1)
3
)
, Uk
(
ζ
(β,2)
3
)
,
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·
Uk
(
ζ
(β,1)
k−1
)
, Uk
(
ζ
(β,2)
k−1
)
, . . . , Uk
(
ζ
(β,k−2)
k−1
)
;
here, for any Uk(ζ
(β,h)
i ), we have β
(h)
1 + 1 ≤ β ≤ m. We now show that these equations vanish.
The proof is by straightforward calculation, though tedious and requiring a case distinction. Let
1 < i < k. Fix some β
(h)
1 + 1 ≤ β ≤ m. Consider the equation
Ui
(
ζ
(β,h)
k
)
= Uk
(
ζ
(β,h)
i
)
. (59)
Then h < i < k. According to the structure of the vector ζi, the entries of which in its hth block
are of two kinds, there are two cases.
1. The interrelation for ζ
(β,h)
i is an equality: ζ
(β,h)
i = ζ
(β,i)
h . This is so if, and only if,
β
(i)
1 + 1 ≤ β ≤ m according to the structure of ζi. Now there arise two subcases.
(a) The other interrelation is an equality, too: ζ
(β,h)
k = ζ
(β,k)
h . This is so if, and only if,
β
(k)
1 + 1 ≤ β ≤ m according to the structure of ζk. In this subcase, equation (59)
becomes
Ui
(
ζ
(β,k)
h
)
= Uk
(
ζ
(β,i)
h
)
. (60)
Since the system (34) contains the equalities Ui
(
ζ
(β,k)
h
)
= Uh
(
ζ
(β,k)
i
)
and Uk
(
ζ
(β,i)
h
)
=
Uh
(
ζ
(β,i)
k
)
, equation (60) becomes
Uh
(
ζ
(β,k)
i
)
= Uh
(
ζ
(β,i)
k
)
.
Since i < k and β
(k)
1 + 1 ≤ β ≤ m, from the structure of ζk follows ζ
(β,k)
i = ζ
(β,i)
k .
Thus, equation (59) vanishes.
(b) The other interrelation is an affine-linear combination:
ζ
(β,h)
k =
k∑
a=1
m∑
γ=β
(a)
1 +1
Ca
γ
(
φβ
k
)
ζ
(γ,a)
h + C
(1)
h
(
φβ
k
)
.
This is so if, and only if, β
(i)
1 + 1 ≤ β ≤ β
(k)
1 according to the structure of ζk. In this
subcase, the term Ui(ζ
(β,h)
k ) in equation (59) becomes
Ui
(
ζ
(β,h)
k
)
=
k∑
a=1
m∑
γ=β
(a)
1 +1
Ca
γ (φβ
k)Ui
(
ζ
(γ,a)
h
)
(61a)
38 D. Fesser and W.M. Seiler
+
k∑
a=1
m∑
γ=β
(a)
1 +1
Ui
(
Ca
γ
(
φβ
k
))
ζ
(γ,a)
h + Ui
(
C
(1)
h
(
φβ
k
))
. (61b)
The term Uk(ζ
(β,h)
i ) in equation (59) becomes
Uk
(
ζ
(β,h)
i
)
= Uk
(
ζ
(β,i)
h
)
= Uh
(
ζ
(β,i)
k
)
= Uh
(
k∑
a=1
m∑
γ=β
(a)
1 +1
Ca
γ
(
φβ
k
)
ζ
(γ,a)
i + C
(1)
i
(
φβ
k
))
=
k∑
a=1
m∑
γ=β
(a)
1 +1
Ca
γ
(
φβ
k
)
Uh
(
ζ
(γ,a)
i
)
(62a)
+
k∑
a=1
m∑
γ=β
(a)
1 +1
Uh
(
Ca
γ
(
φβ
k
))
ζ
(γ,a)
i + Uh
(
C
(1)
i
(
φβ
k
))
; (62b)
here we have the first equality because we are considering the first main case, the
second equality because of the structure of the system (34) and the third equality
according to the structure of ζk, since i < k and because β
(i)
1 + 1 ≤ β ≤ β
(k)
1 .
Substituting (61) and (62) in equation (59) and factoring out, we get
0 =
k∑
a=1
m∑
γ=β
(a)
1 +1
Ca
γ (φβ
k)
{
Ui
(
ζ
(γ,a)
h
)
− Uh
(
ζ
(γ,a)
i
)}
(63a)
+
k∑
a=1
m∑
γ=β
(a)
1 +1
{
Ui
(
Ca
γ
(
φβ
k
))
ζ
(γ,a)
h − Uh
(
Ca
γ
(
φβ
k
))
ζ
(γ,a)
i
}
(63b)
+ Ui
(
C
(1)
h
(
φβ
k
))
− Uh
(
C
(1)
i
(
φβ
k
))
. (63c)
Line (63c) contains the Lie bracket [Ui, C
(1)
h ](φβ
k). According to the structure of the
system (34), the term (63a) vanishes. If the terms (63b) and (63c) vanish, too, then
so does equation (34). Otherwise they form a new algebraic condition for (34), which
can be solved for some function ζ
(β,a)
h . Substituting this function in (34) does not
change the classes or the numbers of the single equations therein. Thus, equation (59)
vanishes.
2. The interrelation for ζ
(β,h)
i is an affine-linear combination:
ζ
(β,h)
i =
i∑
a=1
m∑
γ=β
(a)
1 +1
Ca
γ
(
φβ
i
)
ζ
(γ,a)
h + C
(1)
h
(
φβ
i
)
.
This is so if, and only if, β
(h)
1 + 1 ≤ β ≤ β
(i)
1 according to the structure of ζi. Since we
have h < i < k and β
(i)
1 ≤ β
(k)
1 , according to the structure of ζk the other interrelation is
an affine-linear combination, too:
ζ
(β,h)
k =
i∑
b=1
m∑
δ=β
(b)
1 +1
Cb
δ
(
φβ
k
)
ζ
(δ,b)
h + C
(1)
h
(
φβ
k
)
.
Existence and Construction of Vessiot Connections 39
Thus, equation (59) becomes
0 =
i∑
a=1
m∑
γ=β
(a)
1 +1
Ca
γ
(
φβ
i
)
Uk
(
ζ
(γ,a)
h
)
−
k∑
b=1
m∑
δ=β
(b)
1 +1
Cb
δ
(
φβ
k
)
Ui
(
ζ
(δ,b)
h
)
(64a)
+
i∑
a=1
m∑
γ=β
(a)
1 +1
Uk
(
Ca
γ
(
φβ
i
))
ζ
(γ,a)
h −
k∑
b=1
m∑
δ=β
(b)
1 +1
Ui
(
Cb
δ
(
φβ
k
))
ζ
(δ,b)
h (64b)
+ Uk
(
C
(1)
h
(
φβ
i
))
− Ui
(
C
(1)
h
(
φβ
k
))
. (64c)
In part (64a), the terms Uk(ζ
(γ,a)
h ) and Ui(ζ
(δ,b)
h ) are equal to Uh(ζ(γ,a)
k ) and Uh(ζ(δ,b)
i )
according to the structure of the system (34). Thus, equation (64) becomes
0 =
i∑
a=1
m∑
γ=β
(a)
1 +1
Ca
γ
(
φβ
i
)
Uh
(
ζ
(γ,a)
k
)
−
k∑
b=1
m∑
δ=β
(b)
1 +1
Cb
δ
(
φβ
k
)
Uh
(
ζ
(δ,b)
i
)
(64a′)
+ (64b) + (64c).
Factoring out the vector field Uh in part (64a′), this equals
0 = Uh
(
i∑
a=1
m∑
γ=β
(a)
1 +1
Ca
γ
(
φβ
i
)
ζ
(γ,a)
k −
k∑
b=1
m∑
δ=β
(b)
1 +1
Cb
δ
(
φβ
k
)
ζ
(δ,b)
i
)
(65a)
−
(
i∑
a=1
m∑
γ=β
(a)
1 +1
Uh
(
Ca
γ
(
φβ
i
))
ζ
(γ,a)
k −
k∑
b=1
m∑
δ=β
(b)
1 +1
Uh
(
Cb
δ
(
φβ
k
))
ζ
(δ,b)
i
)
(65b)
+ (64b) + (64c). (65c)
Now one can show (see [6, Corollary 3.3.25 (for j = k)]) that the term (65a) equals
Uh
(
C
(1)
i
(
φβ
k
)
− C
(1)
k
(
φβ
i
))
,
which does not contain any ζ
(γ,a)
k or ζ
(δ,b)
i any more; it is an algebraic expression instead of
the differential expression that it seems to be when written in the form (65a). The other
terms, (65b) and (65c), are algebraic, too. So all of equation (59) has shown to be an
algebraic condition when the interrelations between the entries of the vectors ζh, ζi and ζk,
as noted in equations (35) and (44), are taken into account.
If this new algebraic condition for the system (34) vanishes, equation (59) vanishes. Oth-
erwise, this new algebraic condition given in equation (65) now appears as
0 = Uh
(
C
(1)
i
(
φβ
k
)
− C
(1)
k
(
φβ
i
))
(65a′)
−
(
i∑
a=1
m∑
γ=β
(a)
1 +1
Uh
(
Ca
γ
(
φβ
i
))
ζ
(γ,a)
k −
k∑
b=1
m∑
δ=β
(b)
1 +1
Uh
(
Cb
δ
(
φβ
k
))
ζ
(δ,b)
i
)
(65b)
+
i∑
a=1
m∑
γ=β
(a)
1 +1
Uk
(
Ca
γ
(
φβ
i
))
ζ
(γ,a)
h −
k∑
b=1
m∑
δ=β
(b)
1 +1
Ui
(
Cb
δ
(
φβ
k
))
ζ
(δ,b)
h (64b)
+ Uk
(
C
(1)
h
(
φβ
i
))
− Ui
(
C
(1)
h
(
φβ
k
))
. (64c)
40 D. Fesser and W.M. Seiler
Collecting terms in lines (65a′) and (64c), this yields
0 = (64b) + (65b)
+ Uh
(
C
(1)
i
(
φβ
k
))
− Ui
(
C
(1)
h
(
φβ
k
))
+ Uk
(
C
(1)
h
(
φβ
i
))
− Uh
(
C
(1)
k
(
φβ
i
))
.
The lower line contains the Lie brackets [Uh, C
(1)
i ](φβ
k) and [Uk, C
(1)
h ](φβ
i ). There must be
some non-vanishing summand containing a factor ζ
(γ,a)
k , ζ
(δ,b)
i , ζ
(γ,a)
h or ζ
(δ,a)
h . As we did in
case 1. (b), we solve (64) for this non-vanishing factor and substitute it into the system (34),
which does not change the class of any equation therein. Therefore equation (59) drops
out from the system (34).
Now we have shown that all those equations vanish where the leading derivative is subject
to being substituted through the interrelations concerning the coefficient function ζ
(β,i)
k . In the
system (34∗), these are the equations with the leaders(
ζ
(β,1)
2
)
k
,(
ζ
(β,1)
3
)
k
,
(
ζ
(β,2)
3
)
k
,
· · · · · · · · · · · · · · · · · · · · · · · · · · ·(
ζ
(β,1)
k−1
)
k
,
(
ζ
(β,2)
k−1
)
k
, . . . ,
(
ζ
(β,k−2)
k−1
)
k
;
here 2 ≤ k ≤ n and β
(h)
1 + 1 ≤ β ≤ m. The remaining equations still form an involutive system
(we may enumerate the remaining ζp
i in such a way that no gaps appear) as the considerations
for the system (34) in Proposition 7 apply likewise. Thus we eventually arrive at an analytic
involutive differential equation for the coefficient functions ζk
i which is solvable according to the
Cartan–Kähler theorem (Theorem 1).
Acknowledgements
This work received partial financial support by the European NEST-Adventure grant 5006,
Global Integrability of Field Theories.
References
[1] Anderson I., Kamran N., Olver P.J., Internal, external, and generalized symmetries, Adv. Math. 100 (1993),
53–100.
[2] Arnold V.I., Geometrical methods in the theory of ordinary differential equations, Grundlehren der Mathe-
matischen Wissenschaften, Vol. 250, Springer-Verlag, New York, 1988.
[3] Bryant R.L., Chern S.S., Gardner R.B., Goldschmidt H.L., Griffiths P.A., Exterior differential systems,
Mathematical Sciences Research Institute Publications, Vol. 18, Springer-Verlag, New York, 1991.
[4] Cartan É., Les Systèmes différentiels extérieurs et leurs applications géométriques, Actualités Sci. Ind.,
no. 994, Hermann et Cie., Paris, 1945.
[5] Fackerell E.D., Isovectors and prolongation structures by vessiot’s vector field formulation of partial
differential equations, in Geometric Aspects of the Einstein Equations and Integrable Systems (Schevenin-
gen, 1984), Editor R. Martini, Lecture Notes in Phys., Vol. 239, Springer-Verlag, Berlin, 1985, 303–321.
[6] Fesser D., On Vessiot’s theory of partial differential equations. A geometric approach to constructing
infinitesimal solutions, PhD Thesis, Saarbrücken, 2008.
[7] Fesser D., Seiler W.M., Vessiot connections of partial differential equations, in Global Integrability of Field
Theories – GIFT 2006, Editors J. Calmet, W.M. Seiler and R.W. Tucker, Universitätsverlag Karlsruhe.
Karlsruhe, 2006, 111–134.
Existence and Construction of Vessiot Connections 41
[8] Globke W., Eine Objektorientierte Programmierumgebung für Differentialgeometrische Berechnungen in
MuPAD, Project Thesis, Fakultät für Informatik, Universität Karlsruhe, 2006.
[9] Golubitsky M., Guillemin V.W., Stable mappings and their singularities, Graduate Texts in Mathematics,
Vol. 14, Springer-Verlag, New York – Heidelberg, 1973.
[10] Hartley D.H., Tucker R.W., A constructive implementation of the Cartan–Kähler theory of exterior diffe-
rential systems, J. Symbolic Comput. 12 (1991), 655–667.
[11] Hausdorf M., Seiler W.M., An efficient algebraic algorithm for the geometric completion to involution, Appl.
Algebra Engrg. Comm. Comput. 13 (2002), 163–207.
[12] Ivey T.A., Landsberg J.M., Cartan for beginners: differential geometry via moving frames and exterior
differential systems, Graduate Studies in Mathematics, Vol. 61, American Mathematical Society, Providence,
RI, 2003.
[13] Krasilshchik I.S., Lychagin V.V., Vinogradov A.M., Geometry of jet spaces and nonlinear partial differential
equations, Advanced Studies in Contemporary Mathematics, Vol. 1, Gordon and Breach Science Publishers,
New York, 1986.
[14] Kruglikov B.S., Lychagin V.V., Geometry of differential equations, in Handbook of Global Analysis, Editors
D. Krupka and D. Saunders, Elsevier, Amsterdam, 2008, 725–771.
[15] Lychagin V.V., Homogeneous geometric structures and homogeneous differential equations, in The Interplay
between Differential Geometry and Differential Equations, Editor V.V. Lychagin, Amer. Math. Soc. Transl.
Ser. 2, Vol. 167, Amer. Math. Soc., Providence, RI, 1995, 143–164.
[16] Lychagin V.V., Singularities of multivalued solutions of nonlinear differential equations, and nonlinear phe-
nomena, Acta Appl. Math. 3 (1985), 135–173.
[17] Modugno M., Covariant quantum mechanics, Unpublished manuscript, Dept. of Mathematics, University of
Florence, 1999.
[18] Olver P.J., Applications of Lie groups to differential equations, Graduate Texts in Mathematics, Vol. 107,
Springer-Verlag, New York, 1986.
[19] Pommaret J.F., Systems of partial differential equations and Lie pseudogroups, Mathematics and its Appli-
cations, Vol. 14, Gordon & Breach Science Publishers, New York, 1978.
[20] Saunders D.J., The geometry of jet bundles, London Mathematical Society Lecture Note Series, Vol. 142,
Cambridge University Press, Cambridge, 1989.
[21] Seiler W.M., Completion to involution and semi-discretisations, Appl. Numer. Math. 42 (2002), 437–451.
[22] Seiler W.M., Spencer cohomology, differential equations, and Pommaret bases, in Gröbner Bases and Sym-
bolic Analysis, Editors M. Rosenkranz and D.M. Wang, Radon Ser. Comput. Appl. Math., Vol. 2, Walter
de Gruyter, Berlin, 2007, 169–216.
[23] Seiler W.M., Involution – the formal theory of differential equations and its applications in computer algebra,
Algorithms and Computation in Mathematics, Vol. 24, Springer-Verlag, Berlin – Heidelberg, 2009.
[24] Stormark O., Lie’s structural approach to PDE systems, Encyclopedia of Mathematics and its Applications,
Vol. 80, Cambridge University Press, Cambridge, 2000.
[25] Tuomela J., On singular points of quasilinear differential and differential-algebraic equations, BIT 37 (1997),
968–977.
[26] Tuomela J., On the resolution of singularities of ordinary differential equations, Numer. Algorithms 19
(1998), 247–259.
[27] Vassiliou P.J., Vessiot structure for manifolds of (p, q)-hyperbolic type: Darboux integrability and symmetry,
Trans. Amer. Math. Soc. 353 (2001), 1705–1739.
[28] Vessiot E., Sur une théorie nouvelle des problèmes généraux d’intégration, Bull. Soc. Math. France 52
(1924), 336–395.
1 Introduction
2 The contact structure
3 The formal theory of differential equations
4 The Cartan normal form
5 The Vessiot distribution
6 Flat Vessiot connections
7 The existence theorem for integral distributions
8 The existence theorem for flat Vessiot connections
9 Conclusions
A Proof of Theorem 2
B Proof of Proposition 7
C Proof of Theorem 3
References
|