See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/269465435
Lie-Groups as a Tool for Solving Differential Equations
Conference Paper · September 2012
DOI: 10.13140/2.1.3818.4003
CITATIONS READS
0 1,435
1 author:
Michael Frewer
47 PUBLICATIONS 204 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Turbulence and Symmetries View project
Stability Analysis and Symmetries View project
All content following this page was uploaded by Michael Frewer on 13 December 2014.
The user has requested enhancement of the downloaded file.
Lie-Groups as a Tool for Solving
Differential Equations
Michael Frewer
Chair of Fluid Dynamics,
Mechanical Engineering,
TU Darmstadt,
Germany
Aim
On a physically intuitive level:
To clarify what a Lie-group is
The connection of Lie-groups with symmetries of differential equations
How to systematically determine all point-symmetries of differential
equations
How to exploit symmetries then for various purposes:
Integrability tests, complete integration of ODEs, invariant solutions for
ODEs and PDEs, asymptotic solutions, conservation laws, dynamical
invariants, etc.
Not included: Using Lie-groups for modeling, i.e. constructing dynamical
equations which should admit a certain given set of
symmetries
Outline
Short historical remarks on the motivation and progress of Lie's theory
Developing Lie's theory intuitively by using 1st order ODEs
Construct the formal framework of a Lie-group symmetry analysis for
differential equations in general
Present the algorithm to perform a systematic search for all Lie-point
symmetries within a given differential equation (using CA-software)
Discussing some key applications to demonstrate how to exploit or
utilize these symmetries to gain as much information form the
differential equation as possible
The Historical Setting of Lie-groups (I)
The origins of Lie theory: Galois theory had clarified the relationship
between the solutions of polynomial equations and their symmetries.
Sophus Lie came up with the idea to develop a similar theory for
differential equations and their symmetries which he and coworkers
then successfully built.
→ The theory of Transformations Groups / Lie-groups with finite-
dimensional Lie-algebras emerged
→ However, mathematicians claim that although it is a brilliant and
complete theory, the connection with the origins got somewhat lost.
→ This prompted new research activities, nowadays collectively known
as Differential Galois Theory
The Historical Setting of Lie-groups (II)
Symmetry analysis of differential equations was developed and applied by
Sophus Lie during the period 1872-1899
→ This theory enables to derive solutions of differential equations in a
completely algorithmic way without using any special lucky guesses
→ In his lecture “Differentialgleichungen” in 1891 he noted:
“ The older examinations on ordinary differential equations as found in
standard books are not systematic.
The writers developed special integration theories for
homogeneous differential equations, for linear differential equations,
and other special integrable forms of differential equations.
However, the mathematicians did not realize that these special theories
are all contained in the term infinitesimal transformations, which is
closely connected with the term of a one parametric group. ”
The Historical Setting of Lie-groups (III)
Despite this feature, Lie's approach to differential equations was not
exploited for half a century.
→ Only the abstract theory of Lie groups grew, mainly driven by Hermann
Weyl during the mathematical development of Quantum Mechanics.
(The term “Lie-group” was first coined by Hermann Weyl in 1928)
It was around 1940s with the work of George Birkhoff and Leonid Sedov
on dimensional analysis, that Lie's theory gave relevant results in
concrete applied problems
In 1960s the Russian school around Lev Ovsiannikov began to exploit
systematically the methods of symmetry analysis of differential equations
→ For the first time explicit solutions were constructed of any sort of
problems, even complicated, from mathematical physics
The Historical Setting of Lie-groups (IV)
But the real breakthrough of Lie's approach to differential equations
began with the construction and development of Computer Algebra
Systems (CAS) like Reduce (1960), Maxima (1970), Maple (1980) and
Mathematica (1990)
→ Lie's theory usually involves a lot of cumbersome and tedious
calculations, therefore it's highly prone to errors
→ For example, in looking for symmetries of PDEs, it is common to
handle hundreds of equations to find a single solution
During the last few decades significant progress has been made in Lie's
approach to differential equations, from both the theoretical and the
applied point of view – and an end is not yet in sight...
Key Idea of Lie's Theory
Lie's theory of symmetry analysis of differential equations relies on the
invariance of differential equations under a transformation of
independent and dependent variables:
→ e.g. scalar 1-dim. PDE of 2nd order:
→ Transformation:
→ Transformation induces the transformation of the derivatives:
→ is a symmetry transformation if:
→ If is a solution of PDE , then also
Key Concept: If the transformation forms a continuous group to some
parameter , then its linear expansion term carries all
information of the transformation, i.e. Lie's theory is a linear theory!
Group properties
The set of symmetry operations / transformations on any object has a
very natural mathematical structure, that of a group:
Closedness:
Associativity:
Unique neutral element:
Unique inverse element:
Commutativity is not part
of the axiom system, i.e. its an additional
property which a group can have or not,
e.g. SO(3) is a non-commutative group:
Geometric investigation on 1st order ODEs
Object under investigation: Scalar ODE of 1st order, for
General form:
To avoid intrinsic singularities:
Examples with different geometric invariances in the solution space:
y-translation:
x-translation:
Rotation in (x,y)-plane:
2D-Scaling:
1st Example
General Solution:
Geometric Form of Solution Space :
Symmetry in : y-Shift (having group structure)
1st Example (Orbits)
Action of the group: Orbits
One Orbit: Locally maps a point on a solution curve / trajectory to a
point on another solution curve
All Orbits: Globally maps a solution to another solution continuously
1st Example (Lie-group)
The set of all orbits induces / defines the global group structure in the
solution space of the differential equation, the so-called Lie-group:
In this example the Lie-group is a y-translation group in
Fixing an IC:
fixes the integration constant:
and hence breaks the symmetry:
1st Example (Natural Coordinate System)
The set of all orbits also induces / defines a natural (canonical) coordinate
system in , where the Lie-group then acts as a pure shift along one
coordinate line and as an invariant in the other coordinate line
Note the difference between coordinate system (static) and the Lie-
group (dynamic)
2nd Example
General Solution:
Geometric Form of Solution Space :
Symmetry in : x-Shift (also having group structure)
2nd Example (Orbits & Lie-group)
Orbits / Lie-group:
Fixing an IC will break the symmetry, except for the IC which fixes the
trivial solution: , which an invariant solution (maps into itself)
2nd Example (Natural Coordinate System)
Orbit-induced natural coordinate system in :
3rd Example
General Solution: Non-trivial in given coordinate system (Cartesian), since
no direct separation of variables possible as before
Geometric Form of Solution Space ,
with asymptotic attractor (“steady state”):
Symmetry in : 2D-rotation (also having group structure)
3rd Example (Orbits & Lie-group)
Orbits / Lie-group:
Any fixing of an IC will break the
symmetry, but only for the
“intermediate” states,
while
the asymptotic solution (attractor
manifold, here a circle of radius 1)
stays invariant
3rd Example (Natural Coordinate System)
Orbit-induced natural coordinate system in :
3rd Example (Integrability)
The rotation symmetry in Cartesian coordinates transforms to a
translation (shift) symmetry in polar coordinates, i.e. the Lie group
transforms to its canonical representation:
Identifying as the new independent and as the new dependent
variable, i.e. , the differential equation transforms via the
chain rule to
The result reflects the fact that the
differential equation / solution space admits a symmetry / Lie-group
which acts as a translation symmetry (in the new coordinates)
Separation of variables now possible → DE is integrable
4th Example
General Solution: Determining is not trivial in this regular (Cartesian)
coordinate system, since a direct separation of
variables is again not possible
Symmetry in having group structure: 2D-Scaling
4th Example (Orbits & Lie-group)
Orbits / Lie-group:
Fixing an IC will fully break the
symmetry by forcing
4th Example (Natural Coordinate System)
Orbit-induced natural coordinate system in :
Locality of Group Orbits
Orbits are not intersecting itself and not intersecting each other, i.e. they are uniquely
(up to some normalization constant) determined by its tangent vector field and vice versa.
→ The local (linear) information of the orbits is fully sufficient to reconstruct the complete
global (nonlinear) orbit
4th Example (Integrability)
The 2D-scaling symmetry in regular Cartesian coordinates transforms
to a shift-symmetry in the new scaled coordinates, i.e. the canonical
representation of the Lie group is
In this coordinate system it is advantageous to identify r as the new
independent and s as the new dependent variable → Separation of
variables then leads to an explicit integration in the original variables
Questions and Answers for
Systematization & Generalization
Q1: Can one systemize in finding all Lie groups and hence its
canonical representations for any given ODE of 1st order ?
Q2: Can it be generalized to ODEs of higher order ?
Q3: Can it be even generalized to PDEs ?
A1: Yes, the exploration is highly algorithmic (Lie's algorithm).
For 1st order ODEs only one symmetry is necessary to integrate
the ODE by a separation of variables
A2: N independent symmetries are necessary for integrability of a
N-th order ODE. Either in one step or in N steps
Think of: Each symmetry reduces order → Leads to classification
of ODEs (Lie managed up to order 2)
A3: Algorithm in finding Lie-point symmetries can be generalized for
PDEs, but not the idea of integrability → only special solutions
emerge under reduction.
Formal Definitions and Properties
of Lie-groups (I)
1-parameter Lie-group:
If: is a differentiable one-to-one mapping
is a continuous group (the orbits in each point)
with analytic composition law
Group parameter can be re-parametrized, such that
with: and
The effect:
Formal Definitions and Properties
of Lie-groups (II)
Define scalar operator:
with the infinitesimal generator (vector field),
more precise: tangent vector field of orbits:
Lie's Central Theorem, Version 1: Exponential-Operator representation:
Lie's Central Theorem, Version 2: Initial-Value-ODE representation:
Linear and Local Condition
for Invariance & Shift
A vector/tensor function is invariant under a transformation if
Restriction for invariance:
Restriction for const. Shift:
Less restrictive condition for invariant surface:
Canonical representation of Lie-groups
Lie-group in original variables:
Aim: To find a coordinate transformation such that the Lie-
group transformation reduces to a single constant shift in only
one variable:
Determining set of linear equations
- for the invariant coordinate lines:
- for the shift coordinate line:
Result: In the new CS the symmetry reduces to:
Extension to Multi-Parameter Lie-groups
r-parametric Lie-group:
Lie's Central Theorem:
with the vector-generator:
and the matrix of infinitesimals:
Disadvantage: In general no additive group composition, i.e. not
commutative
Lie-Group vs Lie-Algebra (I)
If the group parameters are all distinct / essential, then the
generator spans a r-dimensional linear vector
space
This vector space can be geometrically understood as a r-dim.
tangent plane at a point on the surface defined by the r-parametric
Lie-group from that point onwards
Walking in the direction on the
surface via corresponds to
walking in the direction
within the tangential plane
Lie-Group vs Lie-Algebra (II)
Definition of Commutator:
Lie's 2nd Central Theorem:
The r-dim. vector space spanned by the Lie-group infinitesimals
additionally admits the algebraic
structure:
with the constant coefficients known as structure constants.
→ Such a linear space is called a Lie-algebra.
Impact: i) Every r-dim. Lie-group induces a r-dim. Lie-algebra
(a structurized linear space), and vice versa
ii) The knowledge of the structure constants is sufficient to
reconstruct the global “non-linear” Lie-group
→ Sufficient to work only with the Lie-algebra.
(I)
3-parametric Lie-group SO(3):
The non-commutivity of the Lie-group SO(3) is encoded in the
structure constants of the Lie-algebra :
Generators:
(II)
1-parametric subgroup SO(3):
Generator:
→ Note that sub-algebra is more simple than sub-group
Lie-Groups of Differential Equations
General PDE-system:
Remarkably, the search of 1-parameter Lie groups of transformations
leaving differential equations invariant usually leads to r-parameter Lie
groups of transformations
Most general 1-parametric Lie-group transformation (in Jet-space):
We will only consider point transformation:
Prolongation to Differential Equations
The differential variables to not transform independently:
By the Chain Rule the transformation of the differential variables is
induced / fixed by the given or wanted point transformation:
The infinitesimals are surely more easy to determine (also closed
chain)
In search for point transformations which leave the differential equation
invariant (symmetries) only the two sets of infinitesimals
and are to be determined
Local Invariance Criterion
for Differential Equations
Let G be a 1-parameter Lie-group point transformation
●
The infinitesimal generator of G
induces the p-th prolongtion generator on the set of differential
variables
The group G is admitted by the PDE-system (normal-form)
if and only if
Solving the Determining System
The set of equations is the
determining system for the infinitesimals
First impression: System is under-determined if
Since the these two infinitesimals do not depend on the differential
variables the q equations generally splits into a large
overdetermined set of equations. This set is linear, since is a linear
operator in and !
Compare to:
The integration constants define a r-parametric group (even infinite-
dimensional if integration functions are involved), although the search
was for 1-parametric Lie-groups.
Discussing the Point-Symmetries
of 3D Navier-Stokes equations
The 3D incompressible NS-equations in the velocity-pressure
representation:
The global / finite Lie-point symmetries:
Question: Could these symmetries have been guessed by physical
reasoning?
3D Navier-Stokes equations in
Unbounded Spatial Domain
Symmetries get broken as soon as a configuration is specified, e.g. in
an unbounded spatial domain if all fields are decaying sufficiently fast
the 3D NS equations can be written as a pure velocity equation:
The Lie-point symmetry is only compatible with this equation if
otherwise we get unbounded solutions
Complete Integration of ODE 1st order
(I)
Ricatti-type equation:
Not directly integrable by separation of variables
Eqn. admits the scaling symmetry:
or in global / finite form:
Canonical coordinate system in which ODE is integrable by separation
of variables:
Invariant coordinate r is determined by:
Shift coordinate s by:
Special solution is sufficient (look for most simplest one):
Complete Integration of ODE 1st order
(II)
By construction the scaling symmetry transforms to shift-symmetry:
The differential equation transforms to a structure which then does not
explicitly depend on the variable s (Invariance under s-shift)
The transformed differential equation can now by construction be
solved by separation of variables. In this case an explicit solution in the
original variables is possible:
General solution:
Complete Integration of ODE 2nd order
(I)
General form:
Consider the specific nonlinear equation:
admitting the 2-dimensional Lie-algebra:
Aim: Determine canonical coordinate system in which ODE is integrable
by separation of variables:
Idea: Canonical coordinate system must be induced by exploiting both
symmetries simultaneously, i.e. “for each order one symmetry”
Difficulty: Fixing the invariant- and the shift-coordinate by one
symmetry must now be compatible with the other symmetry, given by
the structure of its Lie-algebra.
Complete Integration of ODE 2nd order
(II)
E.g. lets choose the invariant- and shift-coordinate via symmetry
then this choice must be compatible with the Lie-Algebra:
r-coordinate: LS:
RS:
r-constraint:
s-coordinate: LS:
RS:
s-constraint:
Complete Integration of ODE 2nd order
(III)
Most simplest solution sufficient, but should be advantageous later for
the explicit integration:
Four equations with two unknowns
Special solution is sufficient (choose again most simplest one):
By construction the two symmetries in the new / canonical coordinate
system transform as:
Complete Integration of ODE 2nd order
(IV)
The ODE transforms to:
Then by separation of variables one gets the intermediate solution
By construction the separation of variables works again, to get:
In the original variables one then only gets an implicit solution:
Full integrability classification
of ODEs 2nd order
For a 2nd order ODE admitting a 2-dimensional
Lie-algebra:
the classification consists of four cases:
Complete Integration of ODE 3rd order
General form:
For a complete integration one needs at least a 3-dim. Lie-algebra:
Constraint: When fixing the invariant- and the shift-coordinate by one
symmetry, say , it must be now compatible with two remaining
symmetries , given by the structure of the Lie-algebra.
Theorem: The compatibility is given if only the Lie-algebra is solvable.
Definition: A n-dimensional Lie-algebra is solvable if there exists a
series of sub-algebras of dimensions
such that is an ideal of , for all :
Integrability Tests (I)
Hénon-Heiles eqn. (motion of star around galactic center in xy-plane):
Alternative representation as a system of 1st order ODEs:
Phase space is 4-dimensional:
The motion / equations are conservative with the Hamiltonian:
Integrability Tests (II)
Liouville Theorem: A conservative (Hamiltonian) system of equations
is integrable, if the corresponding dynamical system in a 2n-dim. phase
space has n independent constants of motion.
Since the Hénon-Heiles system is conservative and 2x2=4-dim. only
two dynamical invariants need to be found to allow for integrability
The first constant of motion / dynamical invariant is the energy itself
, which is the only one for arbitrary parameters
For arbitrary parameters the dynamics is in general energy conserving
but chaotic (non-integrable)
Only for very specific choices of the parameters the system is fully
integrable, i.e. a second dynamical invariant exists, which restricts the
dynamics to be regular (non-chaotic).
Integrability Tests (III)
Case I:
Case II:
Case III:
Integrability Tests (IV)
Question: Can we tell from a symmetry analysis if the Hénon-Heiles
system is integrable or not ?
For arbitrary parameters the only symmetry of the Hénon-Heiles
equations is the translation symmetry in time:
→ for this symmetry the Noether-Theorem is applicable, and induces
the conserved quantity of energy.
E.g. for Case I, the system additionally admits the point-symmetry:
→ i.e. system is integrable,
→ However, for this symmetry the Noether-Theorem is NOT applicable,
since the theorem gives restrictions which are not satisfied here!
→ How to construct the corresponding conserved quantity from
is a difficult research topic ...
Invariant Solutions of PDEs
As discussed before, the idea of a canonical coordinate system also
extends to PDEs, but does not have the same impact as for ODEs
→ The potential of full integrability is not given
→ Point-symmetries for PDEs “only” induce invariant functions, or
equivalently, reductions to a smaller set of independent variables
Example 1: 2D scalar wave equation:
admits the scaling symmetry:
Canonical coordinate system:
with the dependency-identification:
Invariant coordinates:
Shift coordinate:
Invariant Solutions of PDEs
A special (most simplest) solution is:
By construction the scaling symmetry transforms to:
Hence, the PDE transforms / reduces to a PDE independent of
is called the invariant solution of :
An alternative construction of the invariant solution is:
Invariant Solutions of PDEs
Example 2: The stationary incompressible Prandtl boundary layer
equations over a semi-infinite flat plate:
Change to stream-function representation:
Invariant Solutions of PDEs
The equation
admits e.g. the following scaling symmetry:
inducing the invariant solution:
Canonical coordinate system:
→ invariant coordinates:
→ shift coordinate:
Simplest solution:
→ To get the invariant solution is then simply by merging the invariant
coordinates
Invariant Solutions of PDEs
Inserting the invariant solution ansatz into equation
it gets reduced to an ODE in the invariant variable :
Nice and easy reduction for the equation!
→ But, unfortunately the invariant solution ansatz
is incompatible with the BCs, e.g.
Thus the BCs break the scaling symmetry
Invariant Solutions of PDEs
However, the stationary incompressible Prandtl boundary layer
equations allow for a second scaling symmetry which is compatible with
all BCs!
Construction again of canonical coordinate system for this symmetry:
→ invariant coordinates:
→ shift coordinate:
Simplest solution:
Invariant solution ansatz:
Invariant Solutions of PDEs
Inserting this invariant solution ansatz into the equation
it gets reduced to the following ODE in the invariant variable
All boundary conditions
show full compatibility with the invariant solution ansatz:
Because of this, the equations deserve an own name!
View publication stats
