1. Introduction and Contributions
Least squares Lagrangians on Riemannian manifolds and the problem of best approximation of flatness have gained much attention lately [
1], especially when they are involved in optimization problems whose objectives are integral functionals. Combining this theory with decomposable multivariate dynamics [
2], we get new results in differential geometry and global analysis.
Section 1 outlines the ground material regarding PDEs in differential geometry, least squares Lagrangian densities, dual variational principle, Riemannian volume form, and positive definite differential operators.
Section 2 recalls the basic properties of ∇-flatness, introduces the crucial notion of least squares Lagrangian density attached to ∇-flatness and underlines that the Euclidean metrics extremals are stable with respect to conformal changing. In
Section 3 comes the heart of the paper. Detailing the Riemann-flatness, we introduce the least squares Lagrangian density attached to Riemann-flatness and best approximations of Riemann-flatness solutions. Then the non-flatness extremals are analyzed in detail.
Section 4 shows how Ricci-flatness implies a least squares Lagrangian density and best approximations of Riemann-flatness solutions.
Section 5 lists some analogues of least squares Lagrangian density attached to scalar curvature-flatness and confirms again that Einstein PDEs are extremals.
Section 6 proves some inequalities between our least squares densities.
Section 7 underlines that the least squares technique is suitable for solving some problems in differential geometry.
All Lagrangians we use are written in a local, version which is of special interest for geometers and nonlinear analysts. Their explicit formulas reflect the properties usually needed for differential geometric constructions. In order to make the techniques in this paper available for a broad mathematical audience, we have tried to make the article as much self-contained as possible.
1.1. PDEs in Differential Geometry
The behavior of many different systems in nature and science are governed by a PDEs system. Usually such a system is thought of in terms of coordinates in order to prove the existence of solutions or to find concrete ones. However, tensorial PDEs in differential geometry also contain information that is independent of the choice of coordinates. This is actually the most important information, as it is independent of any external structure artificially added to the PDE, and in this sense is genuine. That is why differential geometry is often considered as an “art of manipulating PDEs” [
3,
4,
5,
6,
7,
8,
9,
10].
The most important geometric PDEs are those producing flatness (e.g., connection-flatness, curvature-flatness, Ricci-flatness, scalar curvature-flatness) and those producing constant curvature (
). The connection-flatness PDEs system is non-tensorial, while curvature-flatness, Ricci-flatness, and scalar curvature-flatness PDEs systems are tensorial. Our ideas come also from the papers [
11,
12,
13,
14,
15,
16,
17,
18,
19].
The connection-flatness and the curvature-flatness are interconnected.
In this paper, we present some specific features: (i) introducing those differential geometric structures needed to define and study geometric PDEs (some of them in a manifestly coordinate-independent way); (ii) defining PDEs and their signification within differential geometry and global analysis; (iii) developing techniques to find intrinsic properties of PDEs; (iv) discussing explicit examples to illustrate the importance of the choice of an appropriate context and language.
1.2. Riemannian Volume Form
Suppose is a smooth oriented Riemannian manifold. Then there is a consistent way to choose the sign of the square root and define a volume form We call it the Riemannian volume form of . Having a volume form allows us to integrate functions on M. In particular, is an important invariant of . It also allows us to define an inner product on the space of differential forms and other tensors or objects on M, using the metric g and its inverse . This inner product induces the square of the norm
1.3. Least Squares Lagrangian Densities
Having in mind the so-called variational approach [
1,
20,
21], in this Subsection we add typical functionals that appear in the theory of geometric and physical fields [
2].
Let M be an oriented manifold of dimension n. Any differential operator (of vectorial form, tensorial or not) on the Riemannian manifold and the metric (geometric structure) g generate a least squares Lagrangian density L. The extremals of the Lagrangian , described by Euler–Lagrange PDEs, include the solutions of initial PDEs and other solutions that we call “Euler–Lagrange prolongations” of those solutions (best approximation of initial PDEs solutions).
Generally, the Euler–Lagrange equation provides the equation of motion for the dynamical field specified in the Lagrangian. If the Lagrangian attached to a PDE is that of the smallest squares, then the extremals give the best approximation of the PDE solutions.
The Euler–Lagrange PDEs are indexed related to the chosen fibered chart
,
. However, since the Euler–Lagrange expressions are components of a global differential form (the Euler–Lagrange form), the solutions are independent of fibered charts [
21].
Example 1. (Compare with the paper [2]) Let be an n-dimensional Riemannian manifold, with local coordinates , and be a compact subset. Let be multi-indices, with each subindex running in . When given Lagrangians , where has multi-components, the associated least squares Lagrangian with respect to the Riemannian metrics , induced by the Riemannian metric , is: The extremals are solutions of the Euler–Lagrange PDE system: If the Lagrangian is associated to the PDEs system , then the extremals contain the solutions of that system and the Euler–Lagrange dynamics are decomposable.
The best example is the least squares Lagrangian of harmonic maps .
Remark 1. If we need to subject the Euler–Lagrange PDEs to boundary conditions, then instead of M we use Ω as a compact, n-dimensional submanifold of M with a boundary (a piece of M).
1.4. Dual Variational Principle
Let be a Riemannian manifold. Usually, the local components of the metric g are denoted by and the components of the inverse are denoted by . Due to the musical isomorphism between the tangent bundle and the cotangent bundle of a Riemannian manifold induced by its metric tensor g, the arbitrary variations of are equivalent to the arbitrary variations of , and any Lagrangian with respect to can be regarded as a Lagrangian in relation to , but the differential orders are different. Consequently, a functional depending on the metric is stationary with respect to variations of the metric if and only if the same functional depending on the metric is stationary with respect to variations of the inverse metric. Noting this fact, we see that we can choose to perform either variations with respect to the metric or its inverse depending on which is more convenient.
When calculating the variation with respect to , certain terms may appear whose integral over any domain can be reduced via the divergence theorem (integration by parts) to an integral over the boundary , which vanish (variations vanish on boundary). Modulo this statement, and the Euler–Lagrange PDEs are reduced to (the formal partial derivatives are equal to zero).
1.5. Positive Definite Differential Operator
For an matrix of numbers or functions, positive definiteness is equivalent to the fact that its leading principal minors are all positive (n inequalities).
For an
matrix of partial derivatives operators, positive definiteness is equivalent to the fact that its leading principal minors are all positive (
n partial differential inequalities). For differential inequalities, see also [
22].
2. Least Squares Lagrangian Density Attached to ∇-Flatness
Let
be a smooth oriented Riemannian manifold. The Riemannian metric
g of components
and its inverse
of components
determine (locally) the Christoffel symbols of the second kind:
(overdetermined elliptic partial differential operator).
From a physical point of view, the Riemannian metric is the gravitational potential and the connection reflects the gravitational field.
The ∇-flatness PDEs system
is
on the space of Riemannian metrics
, i.e.,
consists of distinct first order non-linear nonhomogeneous PDEs whose unknowns are
functions
(positive definite tensor);
indicates an overdetermined system of PDEs;
indicates a determined system. This PDEs system is symmetric in
. Imposing the initial condition
, we find the solution
(Euclidean manifold).
The square of the norm is a Lagrangian density of first order with respect to and of order zero with respect to . The functional describing ∇-flatness deviation is . This can be considered as a functional of g, i.e., , and then we consider variations with respect to g, or a functional of , i.e., , and variations in relation to . Though the second is more simple, from variational point of view, let us begin the study with , whose associated Lagrangian is of first order in .
Theorem 1. The extremals g of , i.e., the solutions of PDEssplit into two categories: (global minimum points, i.e., solutions of ∇
-flatness) and local minimum points of .
Proof. The extremals
of the Lagrangian
are solutions of Euler–Lagrange PDEs:
These critical points are global (when ) or local (when ).
Suppose
. Based on obvious formulas:
we obtain:
The explicit Euler–Lagrange PDEs are those in the Theorem.
Now let us compute the Hessian matrix of components:
This matrix is invariant if one interchanges l with a and the (un-ordered) pair with the (un-ordered) pair , which must happen with a mixed derivative. Since the matrix H is positive and definite, all extremals are minimum points (Legendre–Jacobi criterium). □
2.1. Homothetic Flat Extremals
The extremals g of are Euler–Lagrange prolongations (the best approximations) of the flat solutions . Let us show that the Euclidean metrics extremals are stable with respect to conformal changing.
To simplify the problem, we consider a two-dimensional manifold with the Riemannian metric
. Then the least squares Lagrangian is:
and the Euler–Lagrange PDEs are:
The Lagrangian density
becomes
It follows the Euler–Lagrange PDEs system:
Remark 2. If , then the previous PDEs system is reduced to: Remark 3. If (conformal case), then one gets the PDEs system:which is equivalent to: Since f must be positive throughout, this system of PDEs has only solutions of the form (see Maple (pde, pdsolve(pde)). The metrics with are homothetic to . Consequently, the Euclidean metrics extremals are stable with respect to conformal changing.
For comparison we use , the variations with respect to , and the general form of Euler–Lagrange PDEs (equation of motion for the metric tensor field), and we formulate the following:
Theorem 2. The extremals of are solutions of the PDEs system: For calculus of the matrix
, we need
and
. We find:
This matrix is not definite (neither positive nor negative), since it vanishes in the center of normal coordinates. This is why this matrix is of no help in determining what extremals could be extremum points.
2.2. Homothetic Flatness Extremals
General dimension: Let be a Riemannian manifold of dimension n. Are there are extremals of the type ?
Since
the Euler–Lagrange PDEs are reduced to:
It follows that . Therefore and The metrics with are homothetic to . Consequently, the Euclidean metric extremals are stable with respect to conformal changing.
Bidimensional case: Let us consider a two-dimensional Riemannian manifold with the metric . Let us show again that the Euclidean metric extremals are stable with respect to conformal changing.
In this case:
and the general form of Euler–Lagrange PDEs system is:
Since:
the Euler–Lagrange PDEs system becomes:
Remark 4. If then the Euler–Lagrange PDEs are reduced to , i.e., (Euclidean case).
Remark 5. The conformal case leads to and we get , i.e., f is constant (confirming the general case).
3. Least Squares Lagrangian Density Attached To Riemann-Flatness
Let ∇ be a symmetric connection of components
and
g be a Riemannian metric of components
. We use the operator
, which is a projection, i.e.,
, and is covariantly constant. The Riemann-flatness PDEs system is either
or
where
, and has the general solution
.
Each of the Riemann-flatness PDEs systems is a system of distinct first order linear quadratic PDEs whose unknowns are functions ; indicates an overdetermined system; indicates an undetermined system; indicates a determined system.
The curvature flatness was discussed in [
11,
12,
13,
14,
15,
17] based on the idea of finding an adapted coordinate system. We bring up another point of view, looking for suitable metrics and connections, and not for adapted coordinate systems.
On the smooth oriented manifold , we introduce the square of the norm , which is a Lagrangian density of first order in . It determines a functional (Riemann-flatness deviation) similar to the Yang–Mills functional, namely The extremals ∇ of are solutions of the Euler–Lagrange PDEs (equation of motion for the connection).
Theorem 3. The explicit form of Euler–Lagrange PDEs attached to the Lagrangian is: The Riemann-flatness solutions are global minimum points. The other solutions are best approximation of flatness PDEs solutions.
Let
be a Riemannian manifold. The Riemannian metric
determines the Riemannian curvature tensor field
of components:
where:
In this case Riemannian curvature flatness condition means the tensorial PDEs system
on
, with
distinct second-order linear non-homogeneous PDEs whose unknowns are
functions
(positive definite tensor); for
, undetermined system;
indicates an overdetermined system;
indicates a determined system. This PDEs system is parabolic since the set of eigenvalues of the matrix
(tensorial product of a matrix by itself) contains the eigenvalue 0. Indeed, for all eigenvectors, the eigenvalues are:
-symmetric in
or in
, with
;
-skewsymmetric in
and in
with
. Of course, this PDEs system has all of the properties of a curvature tensor field.
In the general relativity, the Riemann tensor field is a physical observable quantity.
On the Riemannian manifold , we introduce the square of the norm , which is of second order with respect to and of order zero with respect to . The Riemann-flatness deviation is described by the functional . This can be considered as a functional of g, and then we consider variations with respect to g, or a functional of , and finally variations in relation to .
For
the extremals
g are solutions of fourth-order Euler–Lagrange PDEs
while for
the Euler–Lagrange PDEs are reduced to
(equation of motion for the metric tensor field).
Theorem 4. The extremals of the functional are solutions of the PDEs system The Riemann-flat solutions
are global minimum points. So are the metrics obtained from
by changing variables, such as
The other solutions are best approximation of flatness PDEs solutions.
3.1. Non-Flat Extremals
We consider a two-dimensional Riemannian manifold
where
In this case
and
or
The Euler–Lagrange PDEs
become the following system of equations:
Case 1: produces the Euclidean metric.
Case 2:
Equivalently
The conformal case: becomes:
(1) , i.e., Euclidean space.
(2)
Therefore
or
(Poisson PDE). Maple answer (pde, sol: = pdsolve(pde)): this PDE has solutions of the form
, where:
or
Globally, these solutions are not convenient since they are not strictly positive.
We have two particular cases: (a) If
, then
and hence
a Liouville equation with the general solution
This function is strictly positive only locally. (b) If
, then
and hence
a Liouville equation with the general solution
This function is strictly positive only locally.
Theorem 5. The extremals of the Lagrangianare solutions of the PDEs system The Riemann-flat solutions are global minimum points.
3.2. Non-Flat Extremals
We consider a two-dimensional Riemannian manifold
Then,
where:
Since
the following Euler–Lagrange PDEs follow:
The conformal case:
. The PDE system becomes:
The case produces the trivial solution .
We subtract the second equation from the first one and we get:
Particular cases: (a)
and
The second PDE is equivalent to
or to
The Liouville equation has the solution .
The second PDE is equivalent to
or to
The Liouville equation has the solution .
4. Least Squares Lagrangian Density Attached to Ricci-Flatness
A torsion-free affine connection ∇, of components , is called (locally) equiaffine if locally there is a volume form a (nonvanishing n-form) that is parallel with respect to ∇. An affine connection ∇ with zero torsion is Ricci-symmetric if and only if ∇ is locally equiaffine.
Let
be an equiaffine manifold. The components
of the Ricci tensor field
are obtained by the contraction of the first and third indices of the curvature tensor field
, i.e.,
. Each of the Ricci-flatness PDEs systems
is a system of
distinct first order divergence quadratic tensorial PDEs with
unknown functions
;
indicates an undetermined system;
indicates a determined system. Here
works like a trace between
p and
q, in order to produce a divergence term. This operator is associated to the projection
P. Any divergence PDE represents a conservation law.
Let be a Riemannian metric. On the smooth oriented manifold , let us consider the Lagrangian density (square of the norm, first order in ) and the functional (Ricci-flatness deviation) The Euler–Lagrange PDEs are (equation of motion for the connection).
Theorem 6. Let The extremals of the Lagrangian are solutions of the PDEs system: The Ricci-flat solutions are global minimum points. The other solutions are best approximation of flatness PDEs solutions.
In case that
is a Riemannian manifold, the Ricci tensor field
has the components
where
The Ricci tensor field of a connection derived from a Riemannian metric is always symmetric. In this case, a Ricci-flat manifold
means
distinct PDEs with
unknown functions
, on
. They are special cases of Einstein manifolds, where the cosmological constant vanishes. In Physics, Ricci-flat manifolds represent vacuum solutions to the analogues of Einstein’s equations for Riemannian manifolds of any dimension, with a vanishing cosmological constant. In this context we recall some ideas of Professor Bang-Yen Chen (April 15, 2017): “The Ricci tensor is related to the matter content of the universe via Einstein’s field equation in general relativity theory. It is the part of the curvature of spacetime that determines the degree to which matter will tend to converge or diverge in time”.
Let
be a Riemannian manifold. In this case the Ricci flatness was described in the papers [
16,
18,
19,
23,
24,
25] as locally underlining the difference between an “Euclidean ball” and a “geodesic ball”. Surprisingly, there are Ricci-flatness solutions that are not Riemann-flatness solutions, for example the Schwarzschild solution.
On the Riemannian manifold , let us consider the Lagrangian density (square of the norm), which is of second order in and order zero in . The Ricci-flatness deviation is described by the functional .
For
, the extremals
g are solutions of fourth-order Euler–Lagrange PDEs
(equation of motion for the metric tensor field). To simplify, we work first with
, since the Euler–Lagrange PDEs determined by
are reduced to
Also for simplification we use harmonic coordinates when Ricci’s tensor has an easily readable formula [
25]:
Theorem 7. We fix a harmonic coordinate system. The extremals of the functional are solutions of the PDEs system: The Ricci-flat solutions are global minimum points. The other solutions are best approximation of flatness PDEs solutions.
Theorem 8. We fix a harmonic coordinate system. The extremals of the functional are solutions of the PDEs system: The Ricci-flat solutions are global minimum points. The other solutions are best approximation of flatness PDEs solutions.
5. Least Squares Lagrangian Density
Attached to Scalar Curvature-Flatness
Let ∇ be an equiaffine connection of components and be a Riemannian metric, where . On the manifold , we introduce the functional (total scalar curvature) where , and the Lagrangian is of first order with respect to . The general Euler–Lagrange PDEs are (equation of motion for the connection).
Theorem 9. The Euler–Lagrange PDEs attached to the functional , i.e., to the Lagrangian , are: Proof. Since
,
, and
we obtain the PDEs in the Theorem. □
On a smooth oriented Riemannian manifold , we attach the functional (total scalar curvature) . Here the Lagrangian is of the second order with respect to , and of order zero with respect to . In dimension two, this is a topological quantity, namely the Euler characteristic of the Riemann surface according to the Gauss–Bonnet formula. In an dimension we prefer to write the functional in the form
Theorem 10. The Euler–Lagrange PDEs attached to the functional , , i.e., to the Lagrangian , are Einstein PDEs
Proof. The Euler–Lagrange PDEs are
, where
. On the other hand, we have:
The term is of divergence type, and it has no contribution to the Euler–Lagrange equations. Consequently Finally, we obtain the explicit Euler–Lagrange PDEs as . □
Theorem 11 ([
21]).
The solutions of the problem “ subject to ”, are solutions of Einstein PDEs . Proof. We use the Lagrangian
, where
is a constant multiplier. Taking the variations with respect to
, we obtain:
The hypothesis and implies that R is constant. We replace R, respectively , in and we obtain Consequently, and . □
The exact solutions of Einstein PDEs were discussed many times. In dimension four, there are topological obstructions to the existence of Einstein metrics.
On a smooth oriented Riemannian manifold , we attach a scalar curvature-flatness deviation either as the action or as the action .
Theorem 12. The Euler–Lagrange PDEs attached to functional , i.e., to the Lagrangian (zero order with respect to ), are either or .
Corollary 1. The solutions of PDEs or are Euler–Lagrange prolongations of Euclidean metrics .
Remark 6. The solutions of the problem “ subject to ”, are solutions of Einstein PDEs .
6. Some Inequalities between Previous Least Squares Densities
Let
be a Riemannian manifold of dimension
n. The Weyl tensor field is defined by:
When
, the Weyl tensor field is trivially defined to be 0. Additionally, when
, the Weyl tensor field
vanishes identically, and hence:
Generally, the Weyl tensor field vanishes if and only if the manifold
M of dimension
is locally conformally flat.
Now let us show that the densities , , , satisfy some inequalities.
Theorem 13. If , then , with equality only if . If , then , with equality only if the manifold is conformally flat.
Proof. If
, then
. For
, we have:
We use direct computation or computation based on orthogonal decompositions of
, respectively
, concerning the scalar product
. If
, then
. This is a Pythagorean theorem that shows the linear dependence of some least squares densities and implies the inequality in the theorem. If
, then
. This is a Parallelogram theorem that shows the linear dependence of some least squares densities and implies the inequality in the theorem. □
Remark 7. (i) For n = 3, the Lagrangian density is identically zero. (ii) For , the Lagrangian density is identically zero. (iii) For , the Gauss–Bonnet density produces a null Lagrangian [26] (pp. 382–384) (does not contribute to the equations of motion, (locally) producing a total derivative). 7. Conclusions and Future Work
In this paper least squares Lagrangian densities attached to flatness PDEs on Riemannian manifolds were studied. Almost all Lagrangians used in
Section 3,
Section 4 and
Section 5 are covariant generalizations of the Einstein–Hilbert Lagrangian, and their Euler–Lagrange PDEs involve higher-derivative terms and/or higher non-linearities. Such PDEs could be relevant for the early universe and are also typically predicted by quantum theories of gravity, such as string theory [
26] (pp. 382–384). More generally, a density-coupling scalar curvature norm, Ricci norm, Riemann norm, or Weyl norm is
, where
f is of class
.
The Legendre transform of a first-order regular Lagrangian density is the Hamiltonian . Implicitly, we define the generalized momentum . Additionally, reasons from physics call for introducing the energy-momentum tensor field and reasons from differential geometry call for the d-tensor . All of these mathematical objects will be used in our future papers.
The index form technique facilitates the understanding of the significance of the geometric PDEs and of the Lagrangian densities attached to them using the Riemannian metrics. This paper is the continuation of some ideas presented in the papers [
1,
2].
Some of our results were proven for a wider class of manifolds and as a special case we also reproved well-known results for the Einstein PDEs. In addition, the proposed approach can be used for each least squares Lagrangian on Riemannian setting and is able to give intermediate results, which can be seen as the “best approximation” of solutions of geometric PDEs.
In the Riemannian case we have two facilities: (i) the most important Lagrange-type densities are the squares of the norms of important geometric objects: connection, curvature tensor field, Ricci tensor field, and scalar curvature field; (ii) to obtain the Euler–Lagrange PDEs, we can select alternatively either the variations with respect to the metric g or the variations with respect to the inverse metric .
In light of the above discussion, if one is able to say something about the solution of a PDEs system whose solution is a Riemannian metric or an affine connection, one could perhaps say something interesting about the behavior of the manifold and its structure. Further research into the nature of the geometric extremals (metrics or connections) may yield strong theoretic results for finite dimensional Riemannian manifolds.