Abstract
The main goal of these lectures is to give a brief introduction to application of contact geometry to Monge–Ampère equations.
Access provided by Autonomous University of Puebla. Download chapter PDF
Similar content being viewed by others
1 Introduction
The main goal of these lectures is to give a brief introduction to application of contact geometry to Monge–Ampère equations. These equations have the form
where A, B, C, D and E are functions on independent variables x, y, unknown function \(v=v(x,y)\) and its first derivatives \(v_{x},v_{y}\).
Equations of this type arise in various fields. For example, G. Monge considered such equations in connection with the problem of the optimal transportation of sand or soil. This problem was of great importance for the construction of fortifications. A modern modification of this problem has the applications to mathematical economics, especially in taxations problem (Kantorovich–Monge problem [7]).
J.G. Darboux studied and applied such equations in his lectures on general theory of surfaces [3,4,5]. At that time, geometry was a source of various types of equations. For example, the problem of reconstructing a surface with a given Gaussian curvature K(x, y) is equivalent to solving the following equation:
Nowadays, the number of sources of Monge–Ampère equations has increased. Equations arise in physics, aerodynamics, hydrodynamics, filtration theory, in models of the development of oil and gas fields, in meteorology and so on. Some of these applications will be discussed. On the other hand, as we shall see, the Monge–Ampère equations themselves generate geometric structures. For instance, some hyperbolic equations can be considered as almost product structures, and elliptic ones as almost complex structures.
The class of equations is rather wide and contains all linear and quasi-linear equations as we can see. On the other hand, it is the minimal class that contains quasi-linear equations and that is closed with respect to contact transformations.
This fact was known to Sophus Lie, who applied contact geometry methods to this kind of equations. In this paper, S. Lie posed some classification problems for equations with respect to contact pseudogroup. In particular, he posed the problem of equivalence of equations to the quasi-linear and linear forms. This problem was solved by V.V. Lychagin and V.N. Rubtsov [20] (see also [21]) in symplectic case and by A.G. Kushner [12] in contact case. Conditions when equations can be transformed to equations with constant coefficients by contact transformations were found by D.V. Tunitskii [23]. The problem of classification for mixed type equations was solved by A.G. Kushner [9,10,11].
In 1978, V.V. Lychagin noted that the classical Monge–Ampère equations and its multi-dimensional analogues admit effective description in terms of differential forms on the space of 1-jets of smooth functions [16]. His idea was fruitful, and it generated a new approach to Monge–Ampère equations.
The lectures has the following structure.
The first lecture is an introduction to geometry of 1-jets space. We define 1-jets of scalar functions, Cartan distribution, contact transformations and contact vector fields on the 1-jets space [8, 15].
In the second lecture, we describe V.V. Lychagin approach and an introduction to geometry of the Monge–Ampère equations. We follow papers [16, 17] and books [15, 18].
The third lecture is devoted to contact transformations of the Monge–Ampère equations. We consider examples of such transformations and apply them to construct multivalued solutions. We illustrate this on the example of equation arising in filtration theory of two immiscible fluids (oil and water, for example) in porous media [1].
In the fourth lecture, we study geometrical structures associated with non-degenerated (i.e. hyperbolic and elliptic) equations. We consider also the class of so-called symplectic equations and give a criterion of their linearization by symplectic transformation [18, 19].
The last, fifth lecture is devoted to tensor invariants of the Monge–Ampère equations. We construct here differential 2-forms that generalize the well-known Laplace invariants. We follow the papers [12, 14].
All calculations in these lectures are illustrated in the program Maple. The Maple files can be found on the website d-omega.org.
2 Lecture 1. Introduction to Contact Geometry
2.1 Bundle of 1-Jets
Let M be an n-dimensional smooth manifold, \(C^{\infty }(M)\) be the ring of smooth functions on M and \(T^*_aM\) be the cotangent space at the point \(a\in M\).
Definition 2.1
A 1-jet \([f]^1_a\) of a function \(f\in C^{\infty }(M)\) at the point a is a pair
The set of 1-jets at the point \(a\in M\) of all functions
is a vector space with respect to operations of addition and multiplication by real numbers which are pointwise is defined as
Denote by
the set of 1-jets of all smooth functions \(f\in C^\infty (M)\) at all points \(a\in M\).
This is a smooth manifold of dimension \(2\dim M+1\) with local coordinates \(x_1,\,\dots ,\,x_n\,u,\,p_1,\,\dots ,\, p_n\), where \(x_1,\,\dots ,\,x_n\) are local coordinates on M, \(p_1,\,\dots , p_n\) are the induced coordinates on the cotangent bundle and u is the standard coordinate on \(\mathbb {R}\). In other words, the values of these functions at point \([f]^1_k\in J^1M\) are the following:
These coordinates are called canonical.
In what follows we’ll call \(J^1M\) the manifold of 1-jets, and the projection
the 1-jet bundle .
Any function \(f\in C^\infty (M)\) defines the following map:
where
The image
which is a smooth submanifold of \(J^1M\), is called the 1-graph of the function f.
Consider the following differential 1-form
on the 1-jet space \(J^1M\) which we’ll call Cartan form .
It is easy to check that this form does not depend on a choice of canonical coordinates in \(J^1M\).
This form allows us to separate submanifolds of the form \(\Gamma ^1_f\subset J^1M\) from arbitrary submanifolds of dimension n by observation that
for any \(f\in C^\infty (M)\). Indeed,
On the other hand, if a submanifold \(N\subset J^1M\) is a graph of section \(s:M\longrightarrow J^1M\), i.e. \(\pi _1:N\longrightarrow M\) is a diffeomorphism, and
then one can easily check that \(N=\Gamma ^1_f\) for some smooth function \(f\in C^\infty (M)\).
This observation shows that zeroes of the Cartan form (but not the form itself) is important to distinguish 1-graphs from arbitrary submanifolds in \(J^1M\).
Denote by \(\mathcal {C}\) the 2n-dimensional distribution (Cartan distribution) on \(J^1M\) given by zeroes of the Cartan form:
In the dual way, the Cartan distribution can be defined by vector fields tangent to this distribution. Namely, vector fields
give us a local basis in the module of vector fields tangent to \(\mathcal {C}\). This module will be denoted by \(D(\mathcal {C})\).
Then a submanifold \(N\subset J^1M\) is a graph of a smooth function if and only if
-
1.
N is an integral submanifold of the Cartan distribution and
-
2.
\(\pi _1:N\rightarrow M\) is a diffeomorphism.
Remind that a contact structure on an odd-dimensional manifold K, \(\dim K=2k+1\), consists of 2k-dimensional distribution P on K such that
for any differential 1-form \(\lambda \), such that locally \(P=\ker \,\lambda \).
In our case, we have
and therefore the Cartan distribution defines the contact structure on the manifold of 1-jets \(J^1M\).
2.2 Contact Transformations
A transformation \(\Phi \) of the space \(J^1M\) is called contact, if it preserves the Cartan distribution, i.e.
In terms of the Cartan form, a transformation \(\Phi \) is contact if
for some function \(h_\Phi \), or equivalently
Examples of Contact Transformations
-
1.
Translations:
$$ (x_1,x_2,u,p_1,p_2)\longmapsto (x_1+\alpha _1,x_2+\alpha _2,u+\beta ,p_1,p_2), $$where \(\alpha _{1}\), \(\alpha _2\) and \(\beta \) are constants.
-
2.
The Legendre transformation:
$$ (x_1,x_2,u,p_1,p_2)\longmapsto (p_1, p_2, u-x_1p_1-x_2p_2, -x_1,-x_2). $$ -
3.
Partial Legendre’s transformation:
$$ (x_{1},x_{2}, u, p_{1}, p_{2})\longmapsto (p_{1},x_{2},u-p_{1}x_{1},-x_{1},p_{2}). $$
Infinitesimal versions of contact transformations are contact vector fields.
A vector field X on \(J^1M\) is called contact if its local translation group consists of contact transformations.
It means that
for some function \(\lambda _t\) on \(J^1M\). Here, \(\Phi _t\) are shifts along vector field X.
After differentiating both parts of (2.6) by t at \(t=0\), we get:
The left-hand side of the equation is the Lie derivative \(L_X(\varkappa )\) of the Cartan form in the direction of the vector field X and therefore, we get
where h is a function on \(J^1M\).
Multiplying both parts of the last equation by \(\varkappa \), we get:
In canonical coordinates, each contact vector field has the form
for some function f which is called generating function of the contact vector field. Note that
Maple Code: Main Operation on \(J^{1}\mathbb {R}^2\)
-
1.
Load libraries:
-
2.
Set jet notation, declare coordinates on the manifold M and generate coordinates on the 1-jet space:
-
3.
Generate the Cartan form:
-
4.
Define partial Legendre transformation:
-
5.
Apply this transformation to the Cartan form:
-
6.
Prolongation of transformations from \(J^0M\) to \(J^1M\):
-
7.
Define the contact vector field \(X_f\) with generating function \(f=p_2\):
-
8.
Prolongation of vector fields from the plane \(M=\mathbb {R}^2\) to \(J^1M\):
3 Lecture 2. Geometrical Approach to Monge–Ampère Equations
3.1 Non-linear Second-Order Differential Operators
Following [16], any differential n-form \(\omega \) on \(J^1M\) is associated with the differential operator
which acts in the following way:
where (see formula (2.4))
This construction does not cover all non-linear second-order differential operators, but only a certain subclass of them.
Examples
-
1.
The differential 1-form on \(J^{1}\mathbb {R}\)
$$ \omega =(1-x^2)dp+\left( \lambda u-xp\right) dx, $$where
$$ \lambda =\dfrac{a^2}{b^2}, $$generates the Lissajou differential operator
$$\begin{aligned} \Delta _{\omega }(y)=\left( (1-x^2)y''- xy'+\dfrac{a^2}{b^2} y\right) dx. \end{aligned}$$(2.9)Indeed,
$$\begin{aligned} \Delta _{\omega }(v)&=(1-x^2)d\left( y'\right) +\left( -xy' +\frac{a^2}{b^2}y \right) dx \\&=\left( (1-x^2)y''- xy'+\dfrac{a^2}{b^2} y\right) dx. \end{aligned}$$ -
2.
The differential 2-form on \(J^{1}\mathbb {R}^2\)
$$ \omega =dp_{1}\wedge dp_{2}$$generates the Hesse operator
$$\begin{aligned} \Delta _{\omega }(v)=\left( \det {\text {*}}{Hess}v\right) dx_{1}\wedge dx_{2}. \end{aligned}$$(2.10)Indeed,
$$\begin{aligned} \Delta _{\omega }(v)&=d\left( v_{x_{1}}\right) \wedge d\left( v_{x_{2}}\right) \\&=\left( v_{x_{1}x_{1}}dx_{1}+v_{x_{1}x_{2}}dx_{2}\right) \wedge \left( v_{x_{2}x_{1}}dx_{1}+v_{x_{2}x_{2}}dx_{2}\right) \\&=\left( v_{x_{1}x_{1}}v_{x_{2}x_{2}}-v_{x_{1}x_{2}}^{2}\right) dx_{1}\wedge dx_{2}\\&=\left( \det {\text {*}}{Hess}v\right) dx_{1}\wedge dx_{2}, \end{aligned}$$where \({\text {*}}{Hess}v\) is the Hessian of the function v.
-
3.
The differential 3-form
$$\begin{aligned} \omega =p_1dp_1\wedge dx_2\wedge dx_3-dx_1\wedge dp_2\wedge dx_3 - dx_1\wedge dx_2 \wedge dp_3 \end{aligned}$$(2.11)on \(J^{1}\mathbb {R}^{3}\) produces the von Karman differential operator
$$ \left( v_{x}v_{xx}-v_{yy}-v_{zz}\right) dx\wedge dy\wedge dz, $$where \(x=x_1,\, y=x_2,\, z=x_3\).
-
4.
The differential 2-form
$$ \omega =dp_{1}\wedge dx_{2}-dp_{2}\wedge dx_{1}$$on \(J^{1}\mathbb {R}^{2}\) represents the two-dimensional Laplace operator
$$ \Delta _{\omega }(v)=\left( v_{xx}+v_{yy}\right) dx\wedge dy, $$where \(x=x_1,\, y=x_2\).
-
5.
Two differential 2-forms
$$\begin{aligned} \omega =dx_{1}\wedge du\,\qquad \text { and }\qquad \varpi =p_{2}dx_{1}\wedge dx_{2} \end{aligned}$$(2.12)on \(J^{1}\mathbb {R}^{2}\) generate the same operator:
$$\begin{array} [c]{l}\Delta _{\omega }(v)=dx_{1}\wedge \left( v_{x_{1}}dx_{1}+v_{x_{2}}dx_{2}\right) =v_{x_{2}}~dx_{1}\wedge dx_{2},\\ \Delta _{\varpi }(v)=v_{x_{2}}~dx_{1}\wedge dx_{2}. \end{array} $$ -
6.
Any differential n-form
$$\begin{aligned} \omega =\varkappa \wedge \alpha +d\varkappa \wedge \beta \end{aligned}$$(2.13)on \(J^{1}M\), where \(\alpha \in \Omega ^{n-1}\left( J^{1}M\right) \), \(\beta \in \Omega ^{n-2}\left( J^{1}M\right) \) and \(\varkappa \) is the Cartan form, gives the zero operator.
All differential operators \(\Delta _{\omega }\) generate differential equations of second order:
For example, operator (2.9) generates Lissajou equation
Note that the differential operators \(\Delta _{\omega }\) and \(\Delta _{h\omega }\) generate the same equation for each non-zero function h.
Equation (2.14) are called Monge–Ampère equations [16].
The following observation justifies this definition: being written in local canonical contact coordinates on \(J^{1}M\), the operators \(\Delta _{\omega }\) have the same type of non-linearity as the Monge–Ampère equations.
Namely, the non-linearity involves the determinant of the Hesse matrix and its minors. For instance, in the case \(n=2\), for
we get classical Monge–Ampère equations
An advantage of this approach is the reduction of the order of the jet space: we use the simpler space \(J^1M\) instead of the space \(J^2M\) where Monge–Ampère equations should be ad hoc as second-order partial differential equations [8].
The differential equation which is associated with a differential n-form \(\omega \) will be denote by \(\mathcal {E}_\omega \):
The following Maple code generates the corresponding differential operator \(\Delta _{\omega }\) for a differential 2-form \(\omega \) on \(J^{1}\mathbb {R}^2\).
Maple Code: \(\omega \longmapsto \Delta _\omega \)
Construct the differential operator \(\Delta \):
Define a differential 2-form:
Apply the differential operator to this differential form \(\omega =dx_1\wedge dp_1-dx_2 \wedge dp_2\):
As a result, we get the differential operator
3.2 Multivalued Solutions of Monge–Ampère Equations
Let v be a classical solution of the Monge–Ampère equation \(\mathcal {E}_\omega \), i.e. \(\Delta _\omega (v)=0\). Then
It means that the restriction of the differential form \(\omega \) to 1-graph of the function v is zero:
An n-dimensional submanifold \(L\subset J^1M\) is called a multivalued solution of Monge–Ampère equation if
-
1.
L is an integral manifold of the Cartan distribution, i.e. the restriction of the Cartan form to L is zero: \(\varkappa \mid _L=0\);
-
2.
the restriction of the differential n-form \(\omega \) to L is zero, too: \(\omega \mid _L=0\).
Examples: Multivalued Solutions
-
1.
Parameterized curves
$$ L=\left\{ x=\sin \,bt,\,\, y=\cos \, at,\,\, p=-\frac{a\sin \,at}{b\,\cos \,bt}\right\} $$in the space \(J^1\mathbb {R}\) are multivalued solutions of the Lissajou equation
$$\begin{aligned} (1-x^2)y''- xy'+\dfrac{a^2}{b^2} y=0. \end{aligned}$$(2.18)Indeed, the restriction of the differential 1-form
$$ \omega =(1-x^2)dp+\left( \frac{a^2}{b^2}y-xp\right) dx $$on the curve L is zero. The projections of these curves on the plane \((x,\,y)\) are well-known Lissajou curves (see Figs. 2.1, 2.2).
-
2.
Projections of multivalued solutions of the Monge equation
$$ v_{xx}v_{yy}-v_{xy}^{2}=\left( 1+v_x^2+v_y^2\right) ^2 $$to the space \(\mathbb {R}^3\) with coordinates x, y, v are spheres with radius 1 (see Eq. (2.2).
-
3.
Projections of multivalued solutions of the equation
$$\begin{aligned} v_{xx}v_{yy}-v_{xy}^{2}=0 \end{aligned}$$(2.19)to the space \(\mathbb {R}^3\) with coordinates x, y, v are deployable surfaces.
3.3 Effective Forms
Last two examples (2.12) and (2.13) show that the constructed map
“differential n-forms” \(\rightarrow \) “differential operators”
has a huge kernel.
This kernel consists of differential forms that vanish on any integral manifold of the Cartan distribution. All such forms have form (2.13) (see [15]).
Let’s find a submodule of the module \(\Omega ^2(J^1M)\) of differential 2-forms such that the map is bijective (\({\text {dim}}M=2\)).
Differential 2-form \(\omega \in \Omega ^2(J^1M)\) is called effective if
-
1.
\(X_1 \rfloor \omega =0\);
-
2.
\(\omega \wedge d\varkappa =0\).
Here, \(X_1\) is the contact vector field with generating function 1. In canonical coordinates (2.3)
The first condition means that coordinate representation of \(\omega \) does not contain terms \(du\wedge *\), and therefore \(\omega \ne \varkappa \wedge \alpha \) for some differential 1-form \(\alpha \). Second condition means that \(\omega \ne \beta d\varkappa \), for a function \(\beta \).
The module of effective differential 2-forms will be denoted by \(\Omega ^2_\epsilon (J^1M)\).
There is the projection p which maps module \(\Omega ^2(J^1M)\) to the module \(\Omega ^2(\mathcal {C})\) of “differential forms” on the Cartan distribution.
Namely, define
as follows:
Here, \(\Omega ^2(J^1M)\) and \(\Omega ^2(\mathcal {C})\) are modules of 2-forms on the 1-jet manifold \(J^1M\) and on the Cartan distribution \(\mathcal {C}\) respectively. Remark that
i.e. 2-form \(p(\omega )\in \Omega ^2(\mathcal {C})\).
Theorem 2.1
Any differential 2-form \(\omega \in \Omega ^2(\mathcal {C})\) has the unique representation
where \(\omega _\epsilon \in \Omega ^2_\epsilon (J^1M)\) is an effective 2-form and \(\beta \) is a function.
Proof
In our case, the Cartan distribution \(\mathcal {C}\) is four-dimensional. The exterior differential of the Cartan form is non-degenerated 2-form on each Cartan subspace, i.e. \(d\varkappa _\theta \) is a symplectic structure on \(\mathcal {C}(\theta )\) for any \(\theta \in J^1M\). Therefore, formula
uniquely defines a function \(\beta \). Define now differential form
Since \(\omega _\epsilon \wedge d\varkappa =0\), the form \(\omega _\epsilon \) is effective. \(\square \)
The constructed differential form \(\omega _\epsilon \) is called the effective part of the differential form \(\omega \).
Define the operator
which for any differential 2-form \(\omega \) on the space \(J^1M\) gives its effective part.
It is obvious that differential 2-forms \(\omega \) and \(\mathrm {Eff}(\omega )\) generate the same Monge–Ampère equations.
In canonical coordinates
and any effective differential 2-form has the following representation:
where A, B, C, D and E are smooth functions on \(J^{1}M\). This form corresponds to Eq. (2.17).
The following Maple code contains two procedures which generate effective parts of differential 2-forms.
Maple Code: \(\omega \longmapsto \omega _\epsilon \)
-
1.
Projection of a 2-form to the Cartan distribution:
-
2.
Calculation of effective parts of a differential 2-forms:
4 Lecture 3. Contact Transformations of Monge–Ampère Equations
By the definition, contact transformations preserve the Cartan distribution and multiply the Cartan form \(\varkappa \) by a function (see formula (2.5)).
Therefore, contact transformations do not preserve the contact vector field \(X_1\) in general. Because of this, the image of an effective differential form can be not effective.
Let \(\Phi : J^1M\rightarrow J^1M\) be a contact transformation and \(\omega \) be an effective differential 2-form. Then by the image of differential 2-form \(\omega \), we shall understand the effective differential form \(\mathrm {Eff}(\Phi ^*(\omega ))\).
Two Monge–Ampère equations \(\mathcal {E}_\omega \) and \(\mathcal {E}_\varpi \) are contact equivalent if there exist a contact transformation \(\Phi \) such that \(\varpi =h\mathrm {Eff}(\Phi ^*(\omega ))\) for some function h.
Theorem 2.2
If two equations \(\mathcal {E}_\omega \) and \(\mathcal {E}_\varpi \) are contact equivalent, then their contact transformation maps multivalued solutions of one to multivalued solutions of the other.
Note that, in general, contact transformations do not preserve the class of classical solutions: classical solutions can transform to multivalued solutions and vice versa.
Examples of Linearization of Equations by Contact Transformations
-
1.
The von Karman equation
$$\begin{aligned} v_{x_{1}}v_{x_{1}x_{1}}-v_{x_{2}x_{2}}=0 \end{aligned}$$(2.22)becomes the linear equation
$$\begin{aligned} x_{1}v_{x_{2}x_{2}}+v_{x_{1}x_{1}}=0 \end{aligned}$$(2.23)after Legendre transformation (2.24). The last equation is known as the Triccomi equation.
-
2.
Equation
$$ \det {\text {Hess}}\,v=1 $$is generated by the effective differential 2-form
$$ \omega =dp_{1}\wedge dp_{2}-dx_{1}\wedge dx_{2}. $$After the partial Legendre transformation
$$ \Phi : (x_{1},\;x_{2},\;u,\;p_{1},\;p_{2})\mapsto (p_{1},\;x_{2},\;u-p_{1}x_{1},\;-x_{1},\;p_{2}) $$this form becomes
$$ \omega =dx_{2}\wedge dp_{1}-dx_{1}\wedge dp_{2}, $$and corresponds to the Laplace equation
$$\begin{aligned} v_{x_{1}x_{1}}+v_{x_{2}x_{2}}=0. \end{aligned}$$ -
3.
Quasi-linear equation:
$$ A\left( v_{x},v_{y}\right) v_{xx}+2B\left( v_{x},v_{y}\right) v_{xy}+C\left( v_{x},v_{y}\right) v_{yy}=0. $$This equation is represented by the following effective form:
$$\begin{aligned} \omega =&B\left( p_{1},p_{2}\right) \left( dx_{1}\wedge dp_{1}-dx_{2}\wedge dp_{2}\right) +\,C\left( p_{1},p_{2}\right) dx_{1}\wedge dp_{2}-A\left( p_{1},p_{2}\right) dx_{2}\wedge dp_{1}. \end{aligned}$$After the Legendre transformation
$$\begin{aligned} \Phi :(x_{1},\;x_{2},\;u,\;p_{1},\;p_{2})\mapsto (p_{1},\;p_{2},\;u-p_{1}x_{1}-p_{2}x_{2},\;-x_{1},\;-x_{2},) \end{aligned}$$(2.24)we get the following effective form
$$\begin{aligned} \varphi ^{*}(\omega )&=B\left( -x_{1},-x_{2}\right) \left( dx_{1}\wedge dp_{1}-dx_{2}\wedge dp_{2}\right) +\\&-A\left( -x_{1},-x_{2}\right) dx_{1}\wedge dp_{2}+\,C\left( -x_{1},-x_{2}\right) dx_{2}\wedge dp_{1}, \end{aligned}$$which corresponds to the linear equation:
$$ -A\left( -x_{1},-x_{2}\right) v_{x_{2}x_{2}}+2B\left( -x_{1},-x_{2}\right) v_{x_{1}x_{2}}-\,C\left( -x_{1},-x_{2}\right) v_{x_{1}x_{1}}=0. $$
Example
The following equation arises in filtration theory of two immiscible fluids in porous media [1]:
It is used for finding a strategy to control wavefronts in the development of oil fields.
The corresponding differential 2-form is
where \(x_1=x\), \(x_2=y\). Applying the Legendre transformation
we get the following differential 2-form:
This form corresponds to the linear equation
The general solution of the last equation is
where \(F_1\) and \(F_2\) are arbitrary functions. Differentiating both sides of (2.27), we get
Thus, solution (2.27) generate a surface \(L \subset J^1M\):
Applying the inverse Legendre transformation
to L, we get multivalued solutions of equation (2.25) in parametric form (Fig. 2.3):
In order to simplify the last formula, we introduce new parameters
and new functions
In these notation, multivalued solutions of equation (2.25) takes the form:
where k(a) and r(b) are arbitrary functions.
Maple Code: Equation \(u_{xy}-u_xu_{yy}=0\)
Define coordinates on M:
Construct the differential operator \(\Delta \):
Define the differential 2-form \(\omega \):
The Legendre transformation:
Apply the Legendre transformation to \(\omega \):
Construct the differential operator \(\Delta _{\omega _1}\):
Check solution:
Inverse Legendre transformation:
Apply this transformation to the surface L:
As a result, we get formula (2.28).
Check that \(\mathcal {L}\) is a multivalued solution of equation (2.25), i.e. \(\omega \mid \mathcal {L}=0\):
Visualization of the multivalued solution \(\mathcal {L}\):
5 Lecture 4. Geometrical Structures
5.1 Pfaffians
First of all, we remark that the restriction of the differential 2-form \(d\varkappa \) on the Cartan distribution
defines a symplectic structure on Cartan space \(\mathcal {C}(\theta )\subset T_\theta (J^1M)\).
Using this structure and an effective 2-form \(\omega \in \Omega ^2_\epsilon (J^1M)\) we define function \({\text {*}}{Pf}(\omega )\), called Pfaffian, in the following way [20]:
This is a correct construction because \(\omega \wedge \omega \) and \(\Omega \wedge \Omega \) are 4-forms on the four-dimensional Cartan distribution.
In the case when
we get
We say that the Monge–Ampère equation \(\mathcal {E}_\omega \) is hyperbolic, elliptic or parabolic at a domain \(\mathcal {D}\subset J^1M\) if the function \({\text {*}}{Pf}(\omega )\) is negative, positive or zero at each point of \(\mathcal {D}\), respectively.
If the Pfaffian changes the sign in some points of \(\mathcal {D}\), then the equation \(\mathcal {E}_\omega \) is called a mixed type equation (see [10]).
The hyperbolic and elliptic equations are called non-degenerate.
Maple Code: Pfaffian
For example, the Pfaffian of the differential 2-form
which corresponds to wave equation \(u_{xy}=0\) is equal to \(-1\), and as we know this equation is hyperbolic.
The Pfaffian of the differential 2-form
which corresponds to Laplace equation \(u_{xx}+u_{yy}=0\) is equal to 1. Indeed,
5.2 Fields of Endomorphisms
The standard linear algebra allows us to construct a field of endomorphisms
which is associated with an effective 2-form \(\omega \). Here \(D(\mathcal {C})\) is the module of vector fields tangent to \(\mathcal {C}\).
Namely, the 2-form \(\Omega \) is non-degenerated on \(\mathcal {C}\) and the operator \(A_{\omega }\) is uniquely determined by the following formula [19]:
for all vector fields X tangent to \(\mathcal {C}\).
Proposition 2.1
Operators \(A_{\omega }\) satisfy the following properties:
-
1.
\(\Omega (A_{\omega }X,X)=0\).
-
2.
\(\Omega (A_{\omega }X,Y)=\Omega (X,A_{\omega }Y)\).
Proof
-
1.
\(\Omega (A_{\omega }X,X)=\omega (X,X)=0\).
-
2.
\(\Omega (A_{\omega }X,Y)=\omega (X,Y)=-\omega (Y,X)=-\Omega (A_{\omega }Y,X)=\Omega (X,A_{\omega }Y)\).
\(\square \)
Proposition 2.2
The squares of operators \(A_{\omega }\) are scalar and
Proof
First of all
Using Proposition 2.1,
Since \(\omega \) is effective, \(\omega \wedge \Omega =0\). Then
i.e. differential 1-forms \(A_{\omega }^2X\rfloor \Omega \) and \(X\rfloor \Omega \) are linearly dependent. Therefore the square of the operator \(A_{\omega }\) is a scalar: \(A_{\omega }^2=\alpha \).
Let \(X\in D(\mathcal {C})\) be an arbitrary vector field. Applying the operators \(A_{\omega }X\rfloor \) and \(X\rfloor \) to both parts of formula (2.29) we get
Then
Suppose that \((A_{\omega }X\rfloor \Omega )\wedge (X\rfloor \Omega )=0\). Then the vector fields X and \(A_{\omega }X\) are linearly dependent. Since X is an arbitrary vector field we see that the operator \(A_{\omega }\) is scalar, i.e. \(A_{\omega }X=\lambda X\) for any X. Then
Therefore \(\omega =\lambda \Omega \), which is impossible. So from (2.33), it follows that \({\text {*}}{Pf}(\omega )+\alpha =0\), i.e. \(A_\omega ^{2}+{\text {*}}{Pf}(\omega )=0\). \(\square \)
Let’s find a coordinate representation of the operator \(A_\omega \). Let
be a local basis of the module \(D(\mathcal {C})\). Then formula (2.31) gives:
in this basis.
Maple Code: Operator \(A_\omega \)
Coordinates on the 1-jet space:
Cartan’s form and its exterior differential:
Define 2-form \(\omega \):
Vector fields and 1-forms on Cartan’s distribution:
Checking duality:
Construct an arbitrary vector field on Cartan’s distribution:
General form of \(A=A_\omega \). Here \(a_{i,j}\) are arbitrary functions:
Action of \(A_\omega \) on vector fields:
Equations with respect to \(a_{i,j}\):
5.3 Characteristic Distributions
Effective forms \(\omega \) and \(h\omega \), where h is any non-vanishing function, define the same Monge–Ampère equation. Therefore, for a non-degenerated equation \(\mathcal {E}_\omega \) the form \(\omega \) can be normed in such a way that \(|{\text {*}}{Pf}(\omega )|=1\). It is sufficient to replace \(\omega \) by
By (2.32), the hyperbolic equations generate a product structure
and elliptic equations generate a complex structure
on the Cartan space \(\mathcal {C}(a)\) [18].
Therefore, a non-degenerated Monge–Ampère equation generates two two-dimensional (complex—for elliptic case) distributions on \(J^1M\), which are eigenspaces of the operator \(A_{\omega }\).
These distributions \(\mathcal {C}_{+}(a)\) and \(\mathcal {C}_{-}(a)\) correspond to the eigenvalues 1 and \(-1\) for the hyperbolic equations or to \(\iota \) and \(-\iota \) for the elliptic ones, respectively. Here \(\iota =\sqrt{-1}\).
The distributions \(\mathcal {C}_{+}\) and \(\mathcal {C}_{-}\) are called characteristic.
The characteristic distributions are real for the hyperbolic equations and complex for the elliptic ones. They are complex conjugate for the elliptic equations.
Proposition 2.3
([18]) 1. The characteristic distributions \(\mathcal {C}_{+}\) and \(\mathcal {C}_{-}\) are skew orthogonal with respect to the symplectic structure \(\Omega \), i.e. \(\Omega (X_+, X_-)=0\) for \(X_\pm \in D(\mathcal {C}_\pm )\).
2. On each of them, the 2-form \(\Omega \) is non-degenerate.
On the other hand, any pair of arbitrary real distributions \(\mathcal {C}_{1,0}\) and \(\mathcal {C}_{0,1}\) on \(J^{1}M\) such that
-
1.
\(\dim \mathcal {C}_{1,0}=\dim \mathcal {C}_{0,1}=2\);
-
2.
\(\mathcal {C}=\mathcal {C}_{1,0}\oplus \mathcal {C}_{0,1}\);
-
3.
\(\mathcal {C}_{1,0}\) and \(\mathcal {C}_{0,1}\) are skew-orthogonal with respect to the symplectic structure \(\Omega \)
determines the operator A. Therefore, a hyperbolic Monge–Ampère equation can be regarded as such pair \(\left\{ \mathcal {C}_{1,0}, \mathcal {C}_{0,1}\right\} \) of distributions.
Maple Code: Characteristic Distributions
Calculation of eigenvalues and eigenvectors of the operator \(A_\omega \):
Find the vector fields from the Cartan distribution
For example, the characteristic distribution \(\mathcal {C}_+\) and \(\mathcal {C}_-\) of operator (2.36) are generated by the following vector fields:
and
5.4 Symplectic Monge–Ampère Equations
Monge–Ampère equation (2.17) is called symplectic if its coefficients A, B, C, D, E do not depend on v.
In this case, the structures described above (effective differential forms, the differential operator \(\Delta _\omega \), field of endomorphisms \(A_\omega \)) can be considered on the four-dimensional cotangent bundle \(T^*M\) instead of the five-dimensional jet bundle \(J^1M\).
Below, we repeat main constructions for the symplectic case.
A smooth function \(f\in C^\infty (M)\) defines a section \(s_f: M\longrightarrow T^*M\) of the cotangent bundle
by the following formula:
Let \(\omega \) be a differential 2-form on \(T^*M\). Define a differential operator
Then equation \(\Delta _\omega (v)=0\) is a symplectic Monge–Ampère equation.
Let \(\Omega \) be the symplectic structure on \(T^*M\). In canonical coordinates \(x_1, x_2, p_1, p_2\) on \(T^*M\)
The differential form \(\omega \) is said to be effective if
Pfaffian \({\text {*}}{Pf}(\omega )\) of the differential 2-form \(\omega \) is defined by the following equality:
and formula
defines the field of endomorphisms \(A_{\omega }\) on \(T^*M\).
The square of operator \(A_{\omega }\) is scalar:
Consider now the case when equation is non-degenerated, i.e. \({\text {*}}{Pf}(\omega )\ne 0\) on \(T^*M\). Then, the operator \(A_\omega \) can be normed (see formula (2.37).
For hyperbolic equations we get almost product structure: \(A_\omega ^2=1\), and for elliptic ones we get almost complex structure: \(A_\omega ^2=-1\).
We say that two symplectic equation \(\mathcal {E}_\omega \) and \(\mathcal {E}_\varpi \) are symplectically equivalent if there exist a symplectic transformation \(\Phi \) such that
for some function h.
The following theorem gives a criterion of symplectic equivalence of non-degenerated Monge–Ampère equation to linear equations with constant coefficients.
Theorem 2.3
([19]) Non-degenerated symplectic Monge–Ampère equation \(\mathcal {E}_\omega \) is symplectically equivalent to wave equation
(in hyperbolic case), or to Laplace equation
(in elliptic case) if and only if the Nijenhuis tensor
where \(A_\omega \) is the normed operator.
Recall that the Nijenhuis tensor \(N_A\) of an operator A is a tensor field of rank (1, 2) given by
for vector fields X and Y.
Condition (2.39) can be written in the following equivalent form [20]:
Maple Code: Symplectic Equation and Nijenhuis Tensor
Below we construct the operator \(A_{\omega }\) for non-linear wave equation
Then we calculate the Nijenhuis tensor \(N_{A_\omega }\) and find conditions under which is this equation symplectically equivalent to the linear wave equation with constant coefficients.
As a result we get
where F is an arbitrary function.
So, Eq. (2.40) is symplectically equivalent to wave equation (2.38) if and only if f is a function in \(x_1\) and \(x_2\) only.
5.5 Splitting of Tangent Spaces
Let us return to the space \(J^1M\).
A non-degenerate equation is called regular if the derivatives \(\mathcal {C}_\pm ^{(k)}\) \((k=1,2,3)\) of the characteristic distributions are constant rank distributions, too.
Below we consider regular equations only. Then, the first derivatives of the characteristic distributions
are three-dimensional. Their intersection
is a one-dimensional distribution, which is transversal to Cartan distribution.
Therefore, for hyperbolic equations, the tangent space \(T_a(J^1M)\) splits into the direct sum (see Fig. 2.4)
at each point \(a\in J^1M\) [18].
For elliptic equations, we get a similar decomposition of the complexification of \(T_a(J^1M)\). In this case, the distribution l is real, too.
6 Lecture 5. Tensor Invariants of Monge–Ampère Equations
6.1 Decomposition of de Rham Complex
Let us construct the decomposition of the de Rham complex, which is generated by the splitting of tangent spaces.
Decomposition (2.41) generates a decomposition of the module of exterior s-forms (or its complexification for elliptic equations). Denote the distributions \(\mathcal {C}_{+},l\), and \(\mathcal {C}_{-}\) by \(P_{1},P_{2}\), and \(P_{3}\), respectively.
Let \(D(J^{1}M)\) be the module of vector fields on \(J^{1}M\), and let \(D_j\) be the module of vector fields tangent to distribution \(P_j\).
Define the following submodules of modules of differential s-forms \(\Omega ^s(J^1M)\):
Then we get the following decomposition of the module of differential s-forms on \(J^1M\):
where \(\mathbf {k=}(k_{1},k_2,k_{3})\) is a multi-index, \(k_{i}\in \left\{ 0,1,\ldots ,\dim P_{i}\right\} \),
and
Three first terms of the decomposition are presented in the diagram (see Fig. 2.5).
The exterior differential also splits into the direct sum
where
Theorem 2.4
([12]) If the multi-index \(\mathbf {t}\) contains one negative component and this component is \(-1\), then the operator \(d_{\mathbf {t}}\) is a \(C^\infty (J^1M)\)-homomorphism, i.e.,
for any function f and any differential form \(\alpha \in \Omega ^\mathbf {k}\).
Due to this theorem, we have the seven homomorphisms, and three of them are zeroes. The non-trivial homomorphisms are the following:
6.2 Tensor Invariants
Consider a case
Then the differential \(d_{\mathbf {t}}\) is a \(C^\infty (J^1M)\)-homomorphism. Note that
if \(q\ne s\). Then, the only non-trivial of \(d_{\mathbf {1}_{j}+\mathbf {1}_{k}-\mathbf {1}_{s}}\) is the restriction to the module \(\Omega ^{\mathbf {1}_{s}}\):
Therefore, the homomorphism \(d_{\mathbf {1}_{j}+\mathbf {1}_{k}-\mathbf {1}_{s}}\) defines a tensor field of the type (2,1). This tensor field we denote by \(\tau _{\mathbf {1}_{j}+\mathbf {1}_{k}-\mathbf {1}_{s}}\):
A unique non-trivial component of this tensor field is its restriction to \(\Omega ^{\mathbf {1}_{s}}\). Note that
coincides with \(d_{\mathbf {1}_{j}+\mathbf {1}_{k}-\mathbf {1}_{s}}\).
Tensor fields \(\tau _{\mathbf {1}_{j}+\mathbf {1}_{k}-\mathbf {1}_{s}}\) are differential invariants of Monge–Ampère equations. So, we get four tensors of (2,1)-type [12]:
Maple Code: Tensor Invariants
Below, we present a program for calculating the tensor \(\tau _{-1, 1, 1}\). The remaining tensors can be found similarly after a small adjustment of the program. In this program, we omit the calculation of the characteristic distributions. They must be calculated in advance (see “Maple Code: Operator \(A_\omega \)” and “Maple Code: Characteristic distributions”).
Construct the distribution l (transversal to the Cartan distribution). We are looking for l as an intersection of derivatives of the characteristic distributions \(C_-^{(1)}\) and \(C_+^{(1)}\). This intersection is one-dimensional and it is generated by the vector field Z which we are looking for.
Basis of the module of vector fields on \(J^1M\) and dual basis:
Decomposition of de Rham complex. Bases of \(\Omega ^1(J^1M)\) and \(\Omega ^2(J^1M)\):
List of elements of the basis of \(\Omega ^2\):
Construct the tensor \(\tau _{-1, 1, 1}\):
Arbitrary differential 2-form:
Arbitrary 2-form from \(\Omega ^{1, 0, 0}\):
Projection of a differential 2-form to \(\Omega ^{0, 1, 1}\):
Example: Hunter–Saxton Equation
Consider the Hunter–Saxton equation
where \(\kappa \) is a constant. This equation is hyperbolic, and it has applications in the theory of liquid crystals [6].
The corresponding effective differential 2-form and the operator \(A_\omega \) are the following:
and
Let’s take the following base in the module of vector fields on \(J^1M\):
The dual basis of the module of differential 1-forms is
The vector fields \(X_1,X_2\) and \(Y_1,Y_2\) form bases in the modules \(D(\mathcal {C}_+)\) and \(D(\mathcal {C}_-)\) respectively. Tensor invariants of Eq. (2.45) have the form
6.3 The Laplace Forms
Define bracket \(\left\langle \alpha \otimes X,\beta \otimes Y\right\rangle \) for decomposable tensors \(\alpha \otimes X\) and \(\beta \otimes Y\) of types (2,1) as follows [12]:
For non-decomposable tensors the bracket is defined by linearity.
Define two differential 2-forms \(\lambda _{-}\) and \(\lambda _{+}\) from the module \(\Omega ^{1,0,1}\) as “wedge contractions” of the tensor fields:
Then tensors (2.46) are called Laplace forms of Monge–Ampère equations \(\mathcal {E}_\omega \).
Example: Laplace Form for Linear Equations
For linear hyperbolic equation
the Laplace forms are
where
are the classical Laplace invariants. This observation justifies our definition.
For linear elliptic equations
Laplace forms generalize Cotton invariants [2].
We emphasize that the classical Laplace invariants (2.49) of Eq. (2.50) are not absolute invariants even with respect to transformations
in contrast to forms \(\lambda _\pm \), which are contact invariants.
Example: Laplace Forms for Hunter–Saxton Equation
The Laplace forms for the Hunter–Saxton equation (2.45) are
6.4 Contact Linearization of the Monge–Ampère Equations
It is well known that if the classical Lagrange invariants h and k of a linear hyperbolic equation is zero, then the equation can be reduced to the wave equation (see [22], for example).
Similar statement is true for the Monge–Ampère equations [14]:
Theorem 2.5
A hyperbolic Monge–Ampère equation is locally contact equivalent to the wave equation
if and only if its Laplace invariants are zero: \(\lambda _+=\lambda _-=0\).
Corollary 2.1
The equation
is locally contact equivalent to the wave equation \(v_{xy}=0\) if and only if the function f has the following form:
where the function \(R=R(x,y,v)\) satisfies to the following ordinary linear differential equation:
Solving this equation we get
where \(\varphi =\varphi (x,y,v)\), \(\Phi =\Phi (v)\), and \(g=g(x,y)\) are arbitrary functions.
The general problem of linearization of non-degenerated Monge–Ampère equations with respect to the contact transformations was solved in [13].
References
Akhmetzianov, A.V., Kushner, A.G., Lychagin, V.V. Integrability of Buckley–Leverett’s filtration model. In: IFAC Proceedings Volumes (IFAC-PapersOnline). 49(12), pp. 1251–1254 (2016)
Cotton, E. Sur les invariants différentiels de quelques équations lineaires aux dérivées partielles du second ordre. In: Ann. Sci. Ecole Norm. Sup. 17, pp. 211–244 (1900)
Darboux, G. Leçons sur la théorie générale des surfaces. Vol. I. Paris, Gauthier-Villars, Imprimeur–Libraire, 1887, vi+514 pp.
Darboux, G. Leçons sur la théorie générale des surfaces. Vol. II. Paris, Gauthier-Villars, 1915, 579 pp.
Darboux, G. Leçons sur la théorie générale des surfaces. Vol. III. Paris, Gauthier-Villars at fils, Imprimeur– Libraire, 1894, viii+512 pp.
Hunter, J.K., Saxton, R. Dynamics of director fields. In: SIAM J. Appl. Math. 51(6), pp. 1498–1521 (1991)
Kantorovich, L. On the translocation of masses. // C.R. (Doklady) Acad. Sci. URSS (N.S.), 37, 199–201 (1942)
Krasil’shchik, I.S., Lychagin, V.V., Vinogradov, A.M. Geometry of jet spaces and nonlinear partial differential equations. Advanced Studies in Contemporary Mathematics, 1, Gordon and Breach Science Publishers, New York, 1986, xx+441 pp.
Kushner, A.G. Classification of mixed type Monge-Ampère equations. In: Pràstaro, A., Rassias, Th.M. (eds) Geometry in Partial Differential Equations, pp. 173–188. World Scientific (1993)
Kushner, A.G. Symplectic geometry of mixed type equations. In: Lychagin, V.V. (ed) The Interplay between Differential Geometry and Differential Equations. Amer. Math. Soc. Transl. Ser. 2, 167, pp. 131–142 (1995)
Kushner, A.G. Monge–Ampère equations and e-structures. In: Dokl. Akad. Nauk 361(5), pp. 595–596 (1998) (Russian). English translation in Doklady Mathematics, 58(1), pp. 103–104 (1998)
Kushner, A.G. A contact linearization problem for Monge–Ampère equations and Laplace invariants. In: Acta Appl. Math. 101(1–3), pp. 177–189 (2008)
Kushner, A.G. Classification of Monge-Ampère equations. In: “Differential Equations: Geometry, Symmetries and Integrability”. Proceedings of the Fifth Abel Symposium, Tromso, Norway, June 17–22, 2008 (Editors: B. Kruglikov, V. Lychagin, E. Straume) pp. 223–256.
Kushner, A.G. On contact equivalence of Monge–Ampère equations to linear equations with constant coefficients. Acta Appl. Math. 109(1), pp. 197–210 (2010)
Kushner, A.G., Lychagin, V.V., Rubtsov, V.N. Contact geometry and nonlinear differential equations. Encyclopedia of Mathematics and Its Applications 101, Cambridge University Press, Cambridge, 2007, xxii+496 pp.
Lychagin, V.V. Contact geometry and nonlinear second-order partial differential equations. In: Dokl. Akad. Nauk SSSR 238(5), pp. 273–276 (1978). English translation in Soviet Math. Dokl. 19(5), pp. 34–38 (1978)
Lychagin, V.V. Contact geometry and nonlinear second-order differential equations. In: Uspekhi Mat. Nauk 34(1 (205)), pp. 137–165 (1979). English translation in Russian Math. Surveys 34(1), pp. 149–180 (1979)
Lychagin, V.V. Lectures on geometry of differential equations. Vol. 1,2. “La Sapienza”, Rome, 1993.
Lychagin, V.V., Rubtsov, V.N. The theorems of Sophus Lie for the Monge–Ampère equations (Russian). In: Dokl. Akad. Nauk BSSR 27(5), pp. 396–398 (1983)
Lychagin, V.V., Rubtsov, V.N. Local classification of Monge–Ampère differential equations. In: Dokl. Akad. Nauk SSSR 272(1), pp. 34–38 (1983)
Lychagin, V.V., Rubtsov, V.N., Chekalov, I.V. A classification of Monge–Ampère equations. In: Ann. Sci. Ecole Norm. Sup. (4) 26(3), pp. 281–308 (1993)
Ovsyannikov, L.V. Group Analysis of Differential Equations. New York, Academic Press, 1982, 416 pp.
Tunitskii, D.V. On the contact linearization of Monge-Ampère equations. In: Izv. Ross. Akad. Nauk, Ser. Matem. 60(2), pp. 195–220 (1996)
Acknowledgements
This work is partially supported by the Russian Foundation for Basic Research (project 18-29-10013). The third author was also supported by the Czech Grant Agency, grant no. GA17-01171S.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2019 Springer Nature Switzerland AG
About this chapter
Cite this chapter
Kushner, A., Lychagin, V.V., Slovák, J. (2019). Lectures on Geometry of Monge–Ampère Equations with Maple. In: Kycia, R., Ułan, M., Schneider, E. (eds) Nonlinear PDEs, Their Geometry, and Applications. Tutorials, Schools, and Workshops in the Mathematical Sciences . Birkhäuser, Cham. https://doi.org/10.1007/978-3-030-17031-8_2
Download citation
DOI: https://doi.org/10.1007/978-3-030-17031-8_2
Published:
Publisher Name: Birkhäuser, Cham
Print ISBN: 978-3-030-17030-1
Online ISBN: 978-3-030-17031-8
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)