STUDIES IN APPLIED MATH 2013-Goncalves-Moving Frames and Conservation Laws For Euclidean Invariant Lagrangians PDF

Moving Frames and Conservation Laws for Euclidean
Invariant Lagrangians
By T. M. N. Gonçalves and E. L. Mansfield
In recent work, the authors show the mathematical structure behind both the
Euler–Lagrange system and the set of conservation laws, in terms of the
differential invariants of the group action and a moving frame. In this paper,
the authors demonstrate that the knowledge of this structure allows to find the
first integrals of the Euler–Lagrange equations, and subsequently, to solve by
quadratures, variational problems that are invariant under the special Euclidean
groups S E(2) and S E(3).
1. Introduction
In 1918, Emmy Noether wrote the seminal paper “Invariante Variationsprobleme”

[1], where she showed that for differential systems derived from a variational
principle, conservation laws could be obtained from Lie group actions that left
the functional invariant.
Recently, in [2], it was proved that for Lagrangians that are invariant under
some Lie symmetry group, Noether’s conservation laws can be written in
terms of a moving frame and vectors of invariants. Furthermore, in [2], the
authors showed that for one-dimensional Lagrangians that are invariant under
a semisimple Lie symmetry group, the new format for the conservation laws
could reduce considerably the calculations needed to solve for the extremal
curves. In particular, a classification was given for variational problems
which are invariant under the three inequivalent S L(2, C) actions on the
Address for correspondence: T. M. N. Gonçalves, School of Mathematics, Statistics and Actuarial

Science, University of Kent, Canterbury CT2 7NF, UK; e-mail: T.M.N.Goncalves@kent.ac.uk
DOI: 10.1111/j.1467-9590.2012.00566.x 134

STUDIES IN APPLIED MATHEMATICS 130:134–166

C 2012 by the Massachusetts Institute of Technology
Moving Frames and Conservation Laws for Euclidean Invariant Lagrangians 135
plane (classified by Lie [3]). This paper is a continuation of [2]; here, the
new structure of Noether’s conservation laws is used to show that solving
one-dimensional variational problems, which are invariant under S E(2) and
S E(3), reduces to solving the Euler–Lagrange equations for the invariants,
which are partial differential equations for these, followed by quadratures. Note
that these groups are not semisimple and arise in many physical applications.
To make this paper self-contained, we start in Section 2 by giving a
brief overview to the theoretical foundations on the result on the structure
of Noether’s conservation laws: the application of moving frames to actions
on jet spaces that produces a symbolic invariant calculus for differential
invariants and their derivatives, and the use of this symbolic invariant calculus
to obtain the Euler–Lagrange equations for variational problems with a Lie
symmetry group directly in terms of the invariants. For the one-dimensional and
multidimensional symmetric variational problems, it was shown, in [4] and [2],
respectively, that Noether’s conservation laws come from a careful collection
of the boundary
terms obtained during the calculation of the first variation
of L [u] = L[u] dx, where L[u] is a smooth function of the independent
variables x, the dependent variables u, and finitely many derivatives of the u α ;
this calculation of the first variation involved two sets of integration by parts.
Here, we demonstrate a small result which simplifies the computation of these
conservation laws.
To illustrate the concepts presented in this section, we use the group
action of S E(2) on the plane as our pedagogical example. Having obtained
the Euler–Lagrange equation, in terms of the curvature invariant, for
one-dimensional variational problems invariant under S E(2) in Section 2, and
its associated Noether’s conservation laws in their new format, we demonstrate
in Section 3 that finding the solution to such variational problems reduces
down to solving the Euler–Lagrange equation for the invariants as functions of
the independent variable followed by quadratures.
In Section 4, we start by computing the invariantized Euler–Lagrange
equations for one-dimensional variational problems symmetric under S E(3)
and its six conservation laws; note that S E(3) is a six-dimensional Lie group.
Then, using the induced group action on the laws, which we show in explicit
detail, we can simplify the above variational problems; their solutions reduce to
solving Euler–Lagrange equations for the invariants followed by quadratures.
2. Structure of Noether’s conservation laws
In this section, we give a brief overview to the theoretical foundations of

our result on the structure of Noether’s conservation laws, namely, concepts
regarding moving frames, differential invariants of a group action, and the
invariant calculus of variations, see Fels and Olver [5, 6], Mansfield [4], and
136 T. M. N. Gonçalves and E. L. Mansfield
Kogan and Olver [7] for further details. We use the S E(2) action on the plane
as our pedagogical example. We follow, in this paper, the notation used in [2]
and [4].
A smooth group acting on a smooth space induces an action on the set of its
smooth curves and surface elements, as well as an action on their higher order
derivatives in the appropriate jet bundle, known as the prolonged curves and
surfaces. In this paper, the set M on which the group G acts is the set of these
prolonged curves and surfaces.
2.1. Moving frames and differential invariants of a group action

Let X be the space of independent variables with coordinates x = (x1 , . . . , x p )
and U the space of dependent variables with coordinates u = (u 1 , . . . , u q ). We
use a multi-index notation to represent the derivatives of u α , e.g.,
∂ |K | u α
u αK = k
,
∂ x1k1 ∂ x2k2 · · · ∂ x pp
where the tuple K = (k1 , . . . , k p ) denotes a multi-index of differentiation
of order |K | = k1 + k2 + · · · + k p . Hence, we represent the coordinates in
M = J n (X × U ), the nth jet bundle, as

z = x1 , . . . , x p , u 1 , . . . , u q , u 11 , . . . .
The operator ∂/∂ xi extends to the total differentiation operator
D ∂ q
∂
Di = = + u αK i α ,
Dxi ∂ xi α=1 K ∂u K
where Di maps J n into J n+1 .

In this paper, we are interested in curves that extremize variational problems
which are invariant under S E(2) and S E(3). Hence, consider a Lagrangian
L(z) dx that is invariant under some symmetry group G. Let the group G act
on the space M as follows:
G × M → M,
(g, z) →
z = g · z,
which satisfies either g · (h · z) = (gh) · z, called a left action, or
g · (h · z) = (hg) · z, called a right action. Then, we say a Lagrangian L(z) dx
is invariant under some group action if
L(z) dx = L(
z) d
x
for all g ∈ G.
O(z)
g
z
Figure 1. A local foliation with a transverse cross section.
Consider a Lie group G acting smoothly on M such that the action is free
and regular. Then, for every z ∈ M, there exists a neighborhood U of z, as
illustrated in Figure 1, such that
- the group orbits have the dimension of the group G and foliate U;
- there is a surface K ⊂ U which intersects the group orbits transversally
at a single point. This surface is called the cross section;
- if O(z) represents the group orbit through z, then the group element
g ∈ G taking z ∈ U to k is unique.
Under these conditions, we can define a right moving frame as the map
ρ : U → G which sends z ∈ U to the unique element ρ(z) ∈ G such that
ρ(z) · z = k, {k} = O(z) ∩ K.
The element g ∈ G in Figure 1 equals ρ(z). Here, we are using moving frames
as reformulated by Fels and Olver [5, 6], in the context of differential algebra.
To obtain the right moving frame, which sends z to k, we must first define
the cross section K as the locus of the set of equations ψ j (z) = 0, j = 1, . . . , r ,
where r is the dimension of G. Normally, the cross section is chosen so as to
ease the calculations. Then, solving the set of equations
ψ j (
z) = ψ j (g · z) = 0, j = 1, . . . , r, (1)
known as the normalization equations, for the r group parameters describing G

yields the right moving frame in parametric form. Hence, the frame obtained
satisfies
ψ j (ρ(z) · z) = 0, j = 1, . . . , r.
By the implicit function theorem, a unique solution of (1) provides an

equivariant map, i.e., for a left action
z) = ρ(z)g −1 ,
ρ(
and for a right action
z) = g −1 ρ(z).
ρ(
Remark 1: As we see later, to obtain the new version of Noether’s

conservation laws, it is necessary that the independent variables are invariant.
If not, then one can reparameterize and set the original independent variables
as depending on the new invariant parameters.
EXAMPLE 1. Consider the S E(2) group acting on curves in the

(x, y(x))-plane as follows:

x cos θ sin θ x −a
= , (2)

y − sin θ cos θ y−b
where θ , a, and b are constants that parameterize the group action. Here, we
are using the inverse action because it simplifies the calculations. Since x is
not invariant, we reparameterize (x, y(x)) as (x(s), y(s)), where s is invariant
and let g ∈ S E(2) act on (x(s), y(s)) as in (2).
There is an induced action on the derivatives y K , where K is the index
of differentiation with respect to s, called the prolonged action. The induced
action on ys is defined to be
d
y d
y/ ds
ys = g · ys =
=
d
s d
s/ ds
by the chain rule, so the action of (2) on ys is

ys = −xs sin θ + ys cos θ.
Similarly,

d2
y 1 d d
y

yss = g · yss = 2
= = −xss sin θ + yss cos θ. (3)
d(
s) d
s/ ds ds d
s
If we consider M to be the space with coordinates (s, x, y, xs , ys , xss , yss , . . .),
then the action is locally free near the identity of S E(2). Thus, taking the
normalization equations to be x = 0,
y = 0, and ys = 0, we obtain

ys
a = x, b = y, and θ = arctan , (4)
xs
as the frame in parametric form.
THEOREM 1. Let ρ(z) be a right moving frame. Then, the quantity

I (z) = ρ(z) · z is an invariant of the group action (see [5]).
Consider z = (z 1 , . . . , z m ) ∈ M and let the normalization equations

z i = ci
for i = 1, . . . , r , where r is the dimension of the group G, then
ρ(z) · z = (c1 , . . . , cr , I (zr +1 ), . . . , I (zm )),
where
I (zl ) = g · zl |g=ρ(z) , l = r + 1, . . . , m.
EXAMPLE 1 (CONT). Evaluating yss given by (3) at the frame (4) yields
xs yss − ys xss
yss |(a=x,b=y,θ=arctan ( xys )) =
.
s
xs2 + ys2
Thus, evaluating
y K at the frame (4) yields a differential invariant.
DEFINITION 1. For any prolonged action in the jet space J n (X × U ), the

invariantized jet coordinates are denoted as
α

Ji = I (xi ) = xi |g=ρ(z) , I Kα = I u αK = u
K g=ρ(z) . (5)
These are also known as the normalized differential invariants.
EXAMPLE 1 (CONT). Evaluating

z = (
s, y, xs ,
x, ys ,
yss ) at the frame (4)
yields
g · z|g=ρ(z) = (
s, y, xs ,
x, ys ,
yss )|g=ρ(z)
y y
=(I (s), I x , I y , I1x , I1 , I11 )
(6)

x s yss − ys x ss
= s, 0, 0, xs2 + ys2 , 0, .
xs2 + ys2
The second, third, and fifth components of (6) correspond to the normalization
equations x = 0, y = 0, and ys = 0, respectively. The fourth and sixth
y
components, I1x and I11 , respectively, are the lowest order differential invariants
and all higher order invariants can be obtained in terms of them and their
derivatives.
THEOREM 2 (REPLACEMENT THEOREM [6]). If f (z) is an invariant, then

f (z) = f (I (z)).
This theorem shows that any invariant can be written in terms of the I (z). It
also allows one to find the I Kα in terms of historically well-known invariants
without having to solve for the frame, if sufficiently many of these are known.
EXAMPLE 1 (CONT). We know that S E(2) preserves |xs |, thus applying

the Replacement Theorem we obtain

x 2 y 2 x
|xs | = xs + ys =
2 2 I1 + I1 = I1 ,

which yields that xs2 + ys2 = I1x up to a sign. Next, we know that the Euclidean
curvature κ is also invariant under S E(2), and we obtain
y y y
xs yss − ys xss I1x I11 − I1 I11
x
I11
κ= 3/2 = x 2 y 2 3/2 = x 2 ,
xs2 + ys2 I1 + I1 I1
y
which gives us I11 in terms of κ and |xs |.
Since we are only considering variational problems with invariant independent

variables, all total differential operators will also be invariant. Hence, the
invariantized differential operators
Di =
Di |g=ρ(z) = Di .
We know that
∂ α
u = u αK i ,
∂ xi K
although the same cannot be said about its invariantized version and in general
Di I Kα = I Kα i .
This leads to the following definition:
DEFINITION 2. Invariant differentiations of the jet coordinates, Ji and I Kα ,

are defined, respectively, as
D j Ji = δi j + Ni j , D j I Kα = I Kα j + M Kα j , (7)
where δi j is the Kronecker delta, and Ni j and M Kα j are the correction terms.
The following theorem provides formulas for the correction terms Ni j and
M Kα j , for which we need to define the following notion of infinitesimal of a
prolonged group action.
Let G be a group parameterized by a1 , . . . , ar , where r = dim(G), in a
neighborhood of the identity element. The infinitesimals of the prolonged group
action with respect to these parameters are

j ∂ xj α ∂ u
α
K
ξi = , φ K ,i = .
∂ai g=e ∂ai g=e
Since ξi and φ Kα ,i are the functions of the xi , for i = 1, . . . , p, u α , for

j
α = 1, . . . , q, and u αK , we can define

ξi (I ) = ξi (J, I β )
j j
and
β
φ Kα ,i (I ) = φ Kα ,i J, I β , I L ,
where the arguments have been invariantized.
THEOREM 3. For a left action on the base space and a right moving frame,
the p × r correction matrix K, which provides the correction terms, is given by
K j = z)|g=ρ(z) = ((Te Rρ )−1 )D j ρ) ,
D j ρ (
where ρ = (ρ1 , . . . , ρr )T is in parameter form and Rρ : G → G is the right
multiplication by ρ. The formulas for the correction terms are

r
r
Ni j = K j ξi (I ), M Kα j = K j φ Kα , (I ),
=1 =1
where is the index for the group parameters and r = dim(G).
The proof of this theorem can be found on page 134 of [4].

The correction matrix K can be computed without explicit knowledge of the
frame; it requires only information on the normalization equations and the
infinitesimals, as shown in the following theorem.
THEOREM 4. Suppose {ψi (z) = 0, i = 1, . . . , r } is the set of normalization

equations. Let the m variables occurring in the normalization equations be
ζ1 , . . . , ζm , where k of these are independent variables, and the remaining m–k
are dependent variables and their derivatives. Set J to be the m × r transpose
of the Jacobian matrix of the left-hand sides of the normalization equations
ψ1 , . . . , ψr , with invariantized arguments,
∂ψ j (I )
Ji j = .
∂ I (ζ j )
Define T to be the invariant p × m total derivative matrix

D
Ti j = I ζj . (8)
Dxi
Moreover, let denote the matrix of infinitesimals with invariantized arguments

∂(g · ζ j )
i j = (I ). (9)
∂ai g=e
Then, the correction matrix K providing the correction terms in the process of
invariant differentiation in Theorem 3 can also be given by
K = −TJ (J)−1 .
We will skip the proof of this theorem, as it can be found on page 136 of [4].
EXAMPLE 2. Consider the S E(2) left action on the (s, t, x(s, t), y(s, t))
space

x cos θ − sin θ x a
s = s,
t = t, = + .

y sin θ cos θ y b
Taking the normalization equations to be x = 0,
y = 0, and ys = 0, then the
ψi are ψ1 (x, y, ys ) = x, ψ2 (x, y, ys ) = y, and ψ3 (x, y, ys ) = ys . Thus, the
arguments of the ψi are x, y, and ys , and the invariantized normalization
y
equations are I x = 0, I y = 0, and I1 = 0.
The following table presents the infinitesimals of the prolonged group action:
s t x y xs ys ···
a 0 0 1 0 0 0 ···
b 0 0 0 1 0 0 ···
θ 0 0 −y x −ys xs ···
Selecting the appropriate columns of the table of infinitesimals, we obtain

x y ys
⎛ ⎞
a 1 0 0
= ⎜ .
b ⎝0 1 0⎟ ⎠
θ 0 0 I1x
The transpose of the Jacobian matrix invariantized, in this case, is equal to

the identity matrix,
ψ1 ψ2 ψ3
⎛x ⎞
I 1 0 0
J= y⎜ ,
I ⎝0 1 0⎟⎠
y
I1 0 0 1
and the invariant total derivative matrix is

x y ys
y
T =s I1x 0 I11 .
t y y
I2x I2 I12
Hence,
a b θ
⎛ y ⎞
I11
−I x 0 − x⎟
K = ⎜ s⎜ 1 I 1 ⎟.
t⎜
⎝ y
y ⎟
I12 ⎠
−I2x −I2 − x
I1
In any case, software exists to calculate these correction terms [8]. Equations (7)
show that the processes of invariantization and differentiation do not commute.
Considering two generating differential invariants I Jα and I Lα and letting
J K = L M so that I JαK = I LαM , then this implies that
D K I Jα − M JαK = D M I Lα − M Lα M . (10)
These equations are called syzygies or differential identities. Note that even
when differential operators commute, there are still nontrivial syzygies between
invariants, as we see in the following example. For more information on
syzygies, see Section 5 in [4].
EXAMPLE 3. Suppose x = x(s, t) and y = y(s, t). Let g ∈ S E(2) act on

(x, y) as in Example 1 and suppose the action leaves s and t invariant. Taking
the normalization equations as before, we obtain
xs xt + ys yt y xs yt − ys xt
xt |g=ρ(z) = I2x = , yt |g=ρ(z) = I2 =
.
xs2 + ys2 xs2 + ys2
Furthermore, since both s and t are invariant, Ds and Dt commute. From
x
Figure 2, we can see that there are two ways in which we can obtain I12 and
since both ways must be equal, we get a syzygy between I2 and η = I1 . The
x x
syzygy is
y
Dt η = Ds I2x − κηI2 . (11)
y y
Similarly, we have a syzygy between I2 and I11 and the syzygy is
y y ηs y y
Dt I11 = Ds2 I2 − Ds I2 − κ 2 η2 I2 + 2κηDs I2x + κs ηI2x . (12)
η
Syzygies will play a crucial role in the obtention of the invariantized
Euler–Lagrange equations and Noether’s conservation laws.
2.2. Invariant calculus of variations

Assume Lagrangians to be smooth functions
of x, u, and finitely many derivatives
of the u α and denote these as L [u] = L[u] dx. Furthermore, suppose these
x x
I22 I122
Dx x
I2x I12
x
I112
Dt
0 I1x x
I11
x
Figure 2. Paths to I12 .
are invariant under some group action and let the κ j , j = 1, . . . , N , denote the
generating differential invariants of that group action. Also, assume that the
action leaves theindependent variables invariant so that the Lagrangians can
be rewritten as L[κ] dx. As mentioned in Remark 1, this can always be
achieved by reparameterization.
To obtain the invariantized Euler–Lagrange equations, Kogan and Olver in
[7] studied invariant calculus of variations from a geometric point of view, we
instead do it from a differential algebra point of view. Thus, to obtain the
Euler–Lagrange equations in terms of the invariants, we proceed in a similar
way as for finding the Euler–Lagrange equations in the original variables
(x, u).
Recall that if x → (x, u) extremizes the functional L [u], then for a small
perturbation of u,

d
0= L [u + εv]
dε ε=0

q
D ∂L
= Eα (L)v α + α
αv + ··· dx
α=1 i
Dx i ∂u i
after differentiation under the integral sign and integration by parts, where
D |K | ∂
Eα = (−1)|K |
K Dx1k1 ···
k
Dx pp ∂u αK
is the Euler operator with respect to the dependent variables u α . The boundary
terms play an important part in the formulas for Noether’s conservation laws,
in the case where the perturbation is given by the group action.
To obtain the invariantized Euler–Lagrange equations, we first introduce

a dummy invariant independent variable t and set the u α = u α (x, t). The
introduction of this new independent variable results in q new invariants
Itα = g · u αt |g=ρ(z) and a set of syzygies Dt κ = HI (ut ) that is
⎛ ⎞ ⎛ 1⎞
κ1 It
⎜ ⎟ ⎜ ⎟
Dt ⎝ ... ⎠ = H ⎝ ... ⎠ , (13)
q
κN It
where H is an N × q matrix of operators depending only on the Di , for

i = 1, . . . , p, the κ j , for j = 1, . . . , N , and their invariant derivatives. Since
all independent variables are invariant, we have that all differential operators
commute, specifically [Di , Dt ] = 0, for all i = 1, . . . , p.
Remark 2: Up to this moment, we have represented the independent

variables as x and the dependent variables as u. Since the examples in this
paper only involve two independent variables, s and t, where t is a dummy
variable introduced to effect the variation, we will represent the coordinates of
the space of dependent variables as x, where the dependent variables will be
the usual space coordinates.
For simplicity, consider a one-dimensional Lagrangian L(x, y, yx , yx x , . . .) dx

with a finite number of arguments, which is invariant under the S E(2)
group action (2). Such variational problems can be rewritten in terms of
the generating invariants of its group action, in this case the Euclidean
curvature, κ, and its derivatives with respect to s, the Euclidean arc length. We
reparameterize (x, y(x)) as (x(s), y(s)), and to fix the parameterization as arc
length, we introduce η = xs2 + ys2 = 1 as a constraint. Thus, we consider the
invariantized variational problem

[L(κ, κs , κss , . . .) − λ(s)(η − 1)] ds, (14)
where λ(s) is a Lagrange multiplier. This constraint does not reduce the
solution set and it will simplify the calculations.
Since symbolically we know that

d D
L [x + εv] = L [x],
dε ε=0 Dt xt =v
mirroring the calculations as for the computation of the Euler–Lagrange

equations in the original variables, we obtain

Dt [L(κ, κs , κss , . . .) − λ(s)(η − 1)] ds

∂L ∂L ∂L 2
= Dt κ + Ds Dt κ + Ds Dt κ + · · · − λ(s)Dt η ds
∂κ ∂κs ∂κss

∂L ∂L ∂L ∂L
= − Ds + Ds 2
− Ds3
+ · · · Dt κ
∂κ ∂κs ∂κss ∂κsss

m−1
∂ L
− λ(s)Dt η + Ds (−1)n Dsn Dsm−1−n Dt κ ds
m=1 n=0
∂κ m

=
Eκ (L) Dt κ − λ(s) Dt η

m−1
∂L
+ Ds (−1) Ds
n n
Dsm−1−n Dt κ ds,
m=1 n=0
∂κm
after differentiating under the integral sign and integrating by parts, where
∂mκ
κm = .
∂s m
Next, we substitute the circled Dt η and Dt κ by their respective syzygies

x

η Ds −κ I2 H1
Dt = y = I (xt ), (15)
κ κs Ds + κ
2 2
I2 H2
where we have already set η = 1; it makes no difference to the final result to
do this at this point. Hence,

(Eκ (L)H2 − λ(s)H1 ) I (xt )

m−1
∂L
+ Ds (−1) Ds
n n
Dsm−1−n Dt κ ds
m=1 n=0
∂κm

(H2∗ (Eκ (L)) − H1∗ (λ(s)))I (xt ) + Ds −λ(s)I2x + Eκ (L)Ds I2
y
=

m−1
∂L
− Ds Eκ (L)I2
y
+ (−1) Ds
n n
Dsm−1−n Dt κ ds, (16)
m=1 n=0
∂κm
after a second set of integration by parts, and where H1∗ and H2∗ are the
respective adjoint operators of H1 and H2 . By definition, we know that I (xt )
contains xt ; hence, from the Fundamental Lemma of Calculus of Variations,
y
the coefficients of I2x and I2 must be zero, that is, they represent, respectively,
the invariantized Euler–Lagrange equations
Ex (L) = κs Eκ (L) + λs ,
E y (L) = Ds2 Eκ (L) + κ 2 Eκ (L) + λ(s)κ.
We use Ex (L) = 0 to eliminate λ(s) instead of E y (L) = 0, as it contains
derivatives of κ of lower order. Since L does not depend on s explicitly, by the
result in page 220 of [4], κs Eκ (L) is a total derivative, specifically
⎛ ⎞

m−1
∂ L
κs Eκ (L) = Ds ⎝ L − (−1) j Dsj κm− j ⎠ ,
m=1 j=0
∂κm
then we obtain
m−1

∂L
λ(s) = −L + (−1) Ds
j j
κm− j , (17)
m=1 j=0
∂κm
where the constant of integration has been absorbed into λ(s) (see Remark
7.1.9. of [4]). Hence, we are left with one invariantized Euler–Lagrange
equation in one unknown
⎛ ⎞

m−1
∂ L
E y (L) = Ds2 Eκ (L) + κ 2 Eκ (L) − κ ⎝L − (−1)m Dsm κm− ⎠
j .
m=1 j=0
∂κm (18)
Equation (18) agrees with the one appearing in Kogan and Olver [7].
We have now covered the theoretical foundations necessary to understand
the structure of Noether’s conservation laws.
2.3. New version of Noether’s conservation laws

If one calculates the conservation laws for one-dimensional variational problems
that are invariant under S E(2) from Noether’s First Theorem (for the formulas
of these, see [9], §5.4, and Proposition 5.98) and then rewrite these in terms of
the Euclidean curvature κ and its derivatives with respect to the Euclidean arc
length s, one obtains
⎛ ⎞ ⎛ ⎞
cos θ − sin θ 0 −λ(s) − κEκ (L)
⎜ ⎟ ⎜ ⎟
⎝ sin θ cos θ 0⎠ ⎝ −Ds Eκ (L) ⎠ = c,

a sin θ − b cos θ a cos θ + b sin θ 1 g=ρ(z)−1 Eκ (L)
(19)
R(g) υ(I )
where ρ(z)−1 is the right moving frame (4), λ(s) is the Lagrange multiplier
obtained in (17), and c the constant vector.
We note that R(g) is the Adjoint representation of S E(2) on its Lie algebra
se(2). To see how part of the calculations in Section 2.2 can yield the result in
(19), we first show how the Adjoint representation is calculated in the context
we will need.
Consider the S E(2) group action, as in Example 2, with generating
infinitesimal vector fields
∂x , ∂y , −y∂x + x∂ y .
Let g ∈ S E(2) act on
v = α∂x + β∂ y + γ (−y∂x + x∂ y ),
where α, β, and γ are constants. Hence,
g · v = α∂x + β∂y + γ (−y∂x + x∂y )
⎛ ⎞⎛ ⎞
cos θ − sin θ 0 ∂x
⎜ ⎟⎜ ⎟
= (α β γ ) ⎝ sin θ cos θ 0⎠ ⎝ ∂y ⎠.
a sin θ − b cos θ a cos θ + b sin θ 1 −y∂x + x∂ y
Thus, R(g) is the Adjoint representation of S E(2), denoted as Ad(g).
Remark 3: In Example 1, we used the right action of S E(2) on the plane

to calculate the right moving frame, as it simplified its calculation. However,
to compute Ad(ρ)−1 , we considered the left action of S E(2) on the plane as in
Example 2, avoiding in this way the need to calculate the inverse of Ad(ρ).
Next, recall the boundary terms, in (16), obtained in the calculation of the
invariantized Euler–Lagrange equations,
−λ(s)I2x + Eκ (L)Ds I2 − (Ds Eκ (L)) I2
y y
m−1
∂L (20)
+ (−1) Ds
n n
Dsm−1−n Dt κ = c,
m=1 n=0
∂κm
y
where c is a constant. Substituting Ds I2 and Dsm−1−n Dt κ for all m in the
above expression by the differential formulas
y y
Ds I2 = I12 − κ I2x ,
y
Dt κ = I112 − 2κ I12
x
,
y y
Ds Dt κ = I1112 − 2κ 2 I12 − 3κ I112
x
− 2κs I12
x
,
..
.
where these were obtained from (7), and rewriting (20) as

⎛ ⎞
−λ(s) − κEκ (L)
⎜ ⎟
⎜−2κ ∂ L − 2κs ∂ L + · · ·⎟
x ⎜⎜ ∂κs ∂κss ⎟
⎟
x
I2 I12 ··· ⎜ ∂L ⎟
⎜ −3κ + ··· ⎟
⎜ ∂κ ⎟
⎝ ss ⎠
..
.

Cx
⎛ ⎞
−Ds Eκ (L)
⎜ ⎟
⎜Eκ (L) − 2κ 2 ∂ L + · · ·⎟
⎜ ⎟
⎜ ∂κss ⎟
⎜ ∂ ⎟
y ⎜ L ⎟
y
+ I2 I12 · · · ⎜ ⎜ + · · · ⎟=c
∂κs ⎟
⎜ ⎟
⎜ ∂L ⎟
⎜ + ··· ⎟
⎜ ∂κss ⎟
⎝ ⎠
..
.

Cy
yields the boundary terms in a form that is linear in the I Kα 2 .

We now let t be a group parameter and define the matrices of invariantized
infinitesimals as

i z i
∂
α
(I ) = ζ j (I ) , ζ j =
i
, (21)
∂a j g=e
where α indexes a dependent variable, zi the dependent variable α and their

derivatives, a j a group parameter, and e the identity element. Note that both
and α (I ), for α = 1, . . . , q, are matrices of invariantized infinitesimals, but
they are not exactly the same matrices; the columns of in (9) correspond
to the variables ζ1 , . . . , ζm appearing in the normalization equations, but the
columns of α (I ) only regard the dependent variable with index α.
If the group parameters are (a1 , . . . , ar ) and t = a j , then from the proof of
Theorem 3 in [2]—which corresponds to Theorem 5—we know that
α α
I2 I12 α
· · · = Ad(ρ)−1 j∗ (I ), (22)
where (Ad(ρ)−1 −1 α α
j∗ ) denotes row j of Ad(ρ) . Substituting the (I2 I12 . . .) in
(21) by (22) for each of the group parameters yields the conservation laws

Ad(ρ)−1 α (I )C α = c
α
in matrix form. Thus, the vector of invariants υ(I ) in (19) equals the sum of
the products of the matrices of invariantized infinitesimals α (I ) with the
vectors C α ,
⎛ ⎞
−λ(s) − κEκ (L)
x (I )C x + y (I )C y = ⎝ −Ds Eκ (L) ⎠ ,
Eκ (L)
where
x xs xss . . . y ys yss ysss ...
⎛ ⎞ ⎛ ⎞
a 1 0 0 ... a 0 0 0 0 ...
(I ) = ⎜
x
, (I ) = ⎜
y
.
b ⎝0 0 0 . . .⎟
⎠ b⎝ 1 0 0 0 ... ⎟
⎠
θ 0 0 −κ ... θ 0 1 0 −κ 2 ...
Finally, Noether’s conservation laws for one-dimensional Lagrangians
invariant under S E(2) can be written as
⎛ ⎞⎛ ⎞
xs −ys 0 −λ(s) − κEκ (L)
⎜ ⎟⎜ ⎟
⎝ ys xs 0⎠ ⎝ −Ds Eκ (L) ⎠ = c, (23)
κ
x ys − yxs x xs + yys 1 E (L)
where xs2 + ys2 = 1 was used to simplify the conservation laws and

m−1
∂L
λ(s) = −L + (−1) Ds
j j
κm− j .
m=1 j=0
∂κm
The following theorem generalizes what we have just seen for one-dimensional
variational problems that are invariant under S E(2); it states that the conservation
laws from Noether’s first theorem can be written as the divergence of the
product of a moving frame with vectors of invariants.

THEOREM 5. Let L(κ1 , κ2 , . . .) dx be invariant under G × M → M,
where M = J n (X × U ), with generating invariants κ j , for j = 1, . . . , N , and
let g · xi = xi , for i = 1, . . . , p. Introduce a dummy variable t to effect the
variation and then integration by parts yields

∂ α α
L(κ1 , κ2 , . . .) dx = E (L)It + Div(P) dx,
∂t α
where this defines the p-tuple P, whose components are of the form

Pi = I Jαt Ci,J
α
, i = 1, . . . , p,
α,J
and the vectors Ciα = (Ci,Jα

). Recall that I Jαt = I (u αJ t ), where J is an index with
respect to the independent variables xi , for i = 1, . . . , p. Let (a1 , . . . , ar ) be
coordinates of G near the identity e, and vi , for i = 1, . . . , r , the associated
infinitesimal vector fields. Furthermore, let Ad(g) be the Adjoint representation
of G with respect to these vector fields. For each dependent variable, define
the matrix of infinitesimals to be

z) = ζji ,
α (
where

z i
∂
ζ ji =
∂a j g=e
are the infinitesimals of the prolonged group action. Let α (I ), for α = 1, . . . , q

be the invariantized version of the above matrices. Then, the r conservation
laws obtained via Noether’s theorem can be written in the form
D
Ad(ρ)−1 υ i (I ) = 0,
i
Dx i
where

υ i (I ) = α (I )Ciα .
α
The proof can be found in [2].

In the process of computing the invariantized Euler–Lagrange equation and
Noether’s conservation laws for variational problems invariant under S E(2),
two sets of integration by parts were performed from which we obtained two
sets of boundary terms. One can verify that the boundary terms

m−1
∂L
(−1) Ds
n n
Dsm−1−n Dt κ
m=1 n=0
∂κm
from the first set of integration by parts do not appear in (23). Following
through the calculations in the proof of Theorem 5 in [2], we show in the
following lemma, in addition, that the first set of boundary terms drops.
LEMMA 1. Let {I Kα } be the set of generating invariants of the prolonged

group action G × M → M, where K is the index of differentiation with respect
to the independent and invariant variables xi , for i = 1, . . . , p. Furthermore,
let t be a dummy invariant independent variable and (a1 , . . . , ar ) the group
parameters of G. Letting t be a group parameter and setting t = a j , for
j = 1, . . . , r , then Dt I Kα equals zero for each group parameter.
Proof : According to Definition 2, the derivative of I Kα is equal to
Dt I Kα = I Kα t + M Kα t ,
which by Theorems 3 and 4 is equivalent to

−J(J)−1 φ αK
Dt I Kα = (Tt∗ ) I Kα t , (24)
1
where (Tt∗ ) corresponds to row t of T, Equation (8). As mentioned in

Equation (22), we know that
γ γ γ γ
It I j1 t I j1 j2 t · · · = Ad(ρ)−1
j∗ (I ), (25)
after we conflate t with a group parameter a j of G, where j = 1, . . . , r and

(Ad(ρ)−1 −1
j∗ ) corresponds to row j of Ad(ρ) . Since Equation (24) can be
rearranged as
γ γ
Dt I Kα = It I j1 t · · · Bγ , (26)
γ
where the components of the vectors B γ are the coefficients of the I γ in (24),
γ γ
we can substitute ( It I j1 t · · ·) on the right-hand side of (26) by (25) for each
group parameter and obtain

Ad(ρ)−1
j∗ γ (I )B γ , j = 1, . . . , r. (27)
γ
Writing the above in matrix form yields

Ad(ρ)−1 γ (I )B γ ,
γ
which is equivalent to

−J(J)−1 φ αK
Ad(ρ) −1
φ αK = Ad(ρ)−1 −φ αK + φ αK = 0.
1
In the next sections, we derive formulas for the extremal curves of

one-dimensional variational problems that are invariant under S E(2) and S E(3).
These formulas will result from the new format of Noether’s conservation laws
presented in this section.
3. Lagrangians invariant under SE(2)
Recall that the invariantized Euler–Lagrange equation for a one-dimensional

Lagrangian invariant under S E(2) is
m−1

2 κ 2 κ ∂L
E (L) = Ds E (L) + κ E (L) − κ L −
y
(−1) Ds
j j
κm− j , (28)
m=1 j=0
∂κm
and its associated conservation laws are

⎛ ⎞⎛ ⎞
xs −ys 0 −λ(s) − κEκ (L)
⎜ ⎟⎜ ⎟
⎝ ys xs 0⎠ ⎝ −Ds Eκ (L) ⎠ = c,
x ys − yxs x xs + yys 1 Eκ (L)

Ad(ρ(z))−1 υ(I )
where
m−1

∂L
λ(s) = −L + (−1) j Dsj κm− j , (29)
m=1 j=0
∂κm
and
∂mκ
κm = .
∂s m
Multiplying both sides of the above conservation laws, Ad(ρ(z))−1 υ(I ) = c,
by Ad(ρ(z)) yields the following system of equations
−λ(s) − κEκ (L) = xs c1 + ys c2 , (30)
−Ds Eκ (L) = xs c2 − ys c1 , (31)
Eκ (L) = yc1 − xc2 + c3 . (32)

Also, we obtain a first integral of the Euler–Lagrange equation as follows.
Define
⎛ ⎞
1 0 0
⎜ ⎟
B = ⎝0 1 0⎠ ,
0 0 0
which satisfies the following equality:
B = Ad(ρ)−T BAd(ρ)−1 .
The first integral of the Euler–Lagrange equation is then
υ T (I )Bυ(I ) = cT Bc,
i.e.,
(λ(s) + κEκ (L))2 + (Ds Eκ (L))2 = c12 + c22 , (33)
with λ(s) as in (29).
Once κ is known, one can verify from the following, that the integration
problem has a straightforward solution. Integrating both sides of Equation (30)
with respect to s yields

xc1 + yc2 = [−λ(s) − κEκ (L)] ds, (34)
and then solving the two linear equations (32) and (34) with respect to x and
y, one obtains

1 κ κ
x(s) = 2 c1 [−λ(s) − κE (L)] ds − c2 E (L) + c2 c3 , (35)
c1 + c22

1 κ κ c22 c3
y(s) = c 2 [−λ(s) − κE (L)] ds + c 1 E (L) + . (36)
c12 + c22 c1
One can easily see that Equation (31) is immediately satisfied as it is the
derivative with respect to s of (32).
Thus, for one-dimensional variational problems invariant under the group
S E(2), to obtain the extremal curves, one only needs to solve (33) for κ and
then use Equations (35) and (36).
EXAMPLE 4. If we consider the Lagrangian with L = 1 in Equations (33),

(35), and (36), then the solution which minimizes the arc length is the equation
of a line, as expected.
EXAMPLE 5. Another famous Lagrangian to consider is

yx2x
5/2 dx,
1 + yx2
which can be rewritten in terms of the invariants as

2
κ − λ(s)(η − 1) ds,
where κ is the Euclidean curvature, λ the Lagrange multiplier, and η − 1 = 0

the constraint introduced to fix parameterization.
Using the equations above, we obtain, as Euler himself did [10], that the
curvature of the minimizing curve satisfies
κss + 12 κ 3 = 0,
or the first integral of the Euler–Lagrange equation obtained above,

4κs2 + κ 4 = c12 + c22 ,
which is solved by an elliptic function, followed by quadratures for x and y,
Equations (35) and (36). Solutions are known as Euler’s elastica. For a good
historical report, see [11].
4. Lagrangians invariant under SE(3)
In the previous section, we showed that for an invariant Lagrangian under S E(2),
the invariantized Euler–Lagrange equation and its associated conservation
laws could be used to solve for the extremal curves. In this section, we will
proceed analogously and present the solution to one-dimensional variational
problems that are invariant under the following S E(3) group action on the
(x(s), y(s), z(s))-space parameterized by the Euclidean arc length s
x = R−1 (x − a),
x → (37)
where
⎛ ⎞
cos β cos γ cos β sin γ sin β
⎜ ⎟
R−1 = ⎝− sin α sin β cos γ − cos α sin γ − sin α sin β sin γ + cos α cos γ sin α cos β ⎠
− cos α sin β cos γ + sin α sin γ − cos α sin β sin γ − sin α cos γ cos α cos β
(38)
represents the rotation in the three-dimensional space, and a = (a, b, c)T , the
translation vector with α, β, γ , a, b, and c as the constants that parameterize
the group action.
To solve S E(3) invariant variational problems, we need to find the element
g ∈ S E(3) that sends the point (x, y, z) to the origin, the tangent to the curve
at the point (x, y, z) to the x-axis, and the normal to the curve at the point
(x, y, z) to the y-axis; in other words, which sends z = (x, y, z, ys , z s , z ss ) to
the cross section (0, 0, 0, 0, 0, 0). For that we solve the normalization equations

x = 0, y = 0, z = 0, ys = 0, zs = 0, and z ss = 0, and thus obtain the right
moving frame

ys (y s z ss − z s yss ) − x s (z s x ss − x s z ss )
a = x, b = y, c = z, α = tan−1 ,
xs2 + ys2 + z s2 (xs yss − ys xss )

−1 z s −1 ys (39)
β = tan , γ = tan .
xs + ys
2 2 x s
Consider a one-dimensional variational problem L [x] that is invariant under

the group action (37). To obtain the invariantized Euler–Lagrange equations,
we first rewrite L [x] in terms of the generating invariants of the group action,
which are the Euclidean curvature

xs × xss
κ= ,

xs
3
and torsion
xsss · (xs × xss )
τ= ,

xs × xss
2
and their derivatives with respect to s.
Sinces representstheEuclideanarclength,theconstraintη = xs2 + ys2 + z s2 =
1 must be introduced into the variational problem to fix parameterization.
Hence, the resulting invariantized functional is

[L(κ, τ, κs , τs , κss , τss , . . .) − λ(s)(η − 1)] ds, (40)
where λ(s) is a Lagrange multiplier. As for S E(2), this constraint does not
reduce the solution set and simplifies the computation of the conservation laws.
Next, we introduce a dummy invariant independent variable t and set
x = x(s, t) to effect the variation. The introduction of a new independent
variable results in three new invariants Itα , for α = x, y, z, and a set of syzygies
⎛ ⎞ ⎛ x⎞
η I2
Dt κ = H I 2 ⎠ ,
⎝ ⎠ ⎝ y
(41)
τ z
I2
where the matrix of operators H is
⎛ ⎞
Ds −κ 0
⎜ κs κ 2 − τ 2 + Ds2 −τs − 2τ Ds ⎟
⎝ ⎠,
τs 3τs 2κs τ τ2 τ 2 κs
τs + 2τ Ds Ds κ + κ − κ 2 Ds + 2τκ Ds2 Ds − κ
+ κ − κ Ds − κ 2 Ds + κ Ds
2 1 3
where we have already set η = 1.

As in Section 2.2, we differentiate (40) with respect to t and then integrate
by parts twice to obtain the invariantized Euler–Lagrange equations
Ex (L) = κs Eκ (L) + τs Eτ (L) − Ds (2τ Eτ (L)) + λs ,

2 κ 2τ 2 τ τs 2τ κs
E (L) = Ds E (L) + Ds E (L) +
y
− 2 Ds Eτ (L) + (κ 2 − τ 2 )Eκ (L)
κ κ κ
+ 2τ κEτ (L) + λ(s)κ,

1 2κs κss τ 2 2κs2

E (L) = − Ds3 Eτ (L) + 2 Ds2 Eτ (L) +
z
+ − 3 − κ Ds Eτ (L)
κ κ κ2 κ κ
− κs Eτ (L) + 2τ Ds Eκ (L) + τs Eκ (L),
and the boundary terms

τ τs τ 2τ τ κ y
(2τ E (L) − λ) I2 +
x
E (L) − Ds E (L) − Ds E (L) I2
κ κ

2τ τ τ2 κs
+ E (L) + E (L) Ds I2 + κEτ (L) − 2τ Eκ (L) − Eτ (L) − 2 Ds Eτ (L)
κ y
κ κ κ
1 1 1
+ Ds2 Eτ (L) I2z − Ds Eτ (L)Ds I2z + Eτ (L)Ds2 I2z = k,
κ κ κ
where k is a constant. Note that by Lemma 1, we can disregard the boundary
terms from the first set of integration by parts.
Using Ex (L) = 0 and the fact that

m−1
∂L
κ τ
κs E (L) + τs E (L) = Ds L − (−1)i Dsi κm−i
m=1 i=0
∂κm

m−1
∂ L
− (−1)i Dsi τm−i
m=1 i=0
∂τm
(see page 220 of [4]), we can eliminate λ. Thus,

m−1
∂L
λ(s) = 2τ Eτ (L) − L + (−1)i Dsi κm−i
m=1 i=0
∂κm
m−1
∂L
+ (−1)i Dsi τm−i , (42)
m=1 i=0
∂τm
where the constant of integration has been absorbed into the Lagrange
multiplier, and hence we obtain two invariantized Euler–Lagrange equations in
two unknowns.
To calculate the conservation laws associated with the invariantized
Euler–Lagrange equations, we must compute the moving frame Ad(ρ)−1 and
the vector of invariants υ(I ). To compute the former, we proceed as in
Section 2.3: we calculate the Adjoint representation Ad(g) of S E(3) with
respect to its generating infinitesimal vector fields
va = ∂x , vb = ∂ y , vc = ∂ z , vα = y∂z − z∂ y ,
vβ = x∂z − z∂x , vγ = x∂ y − y∂x ,
and evaluate it at the frame (39), which yields

ρ F S 0
Ad(ρ(z))−1 = D Xρ ,
F S Dρ F S D
where

xss xs × xss
ρ F S = xs
κ κ
is the Frenet–Serret frame, D is the following diagonal matrix

⎛ ⎞
1 0 0
⎜ ⎟
D = ⎝0 −1 0⎠ ,
0 0 1
and X is the matrix

⎛ ⎞
0 −z y
⎜ ⎟
X =⎝ z 0 −x ⎠ .
−y x 0
As seen in Section 2.3, to obtain the vector of invariants, we must first find
the boundary terms that are linear in the I Kα 2 . To do so, consider the boundary
terms obtained from the calculation of the invariantized Euler–Lagrange
equations

τ τs τ 2τ τ κ y
(2τ E (L) − λ)I2x
+ E (L) − Ds E (L) − Ds E (L) I2
κ κ

2τ τ τ2
+ E (L) + E (L) Ds I2 + κEτ (L) − 2τ Eκ (L) − Eτ (L)
κ y
κ κ

κs 1 1
− 2 Ds Eτ (L) + Ds2 Eτ (L) I2z − Ds Eτ (L)Ds I2z (43)
κ κ κ
1
+ Eτ (L)Ds2 I2z = k.
κ
y
Substituting Ds I2 , Ds I2z , and Ds2 I2z in (43) by the differential formulas
y y
Ds I2 = −κ I2x + I12 + τ I2z ,
y
Ds I2z = −τ I2 + I12
z
,
y y
Ds2 I2z = τ κ I2x − τs I2 − 2τ I12 − τ 2 I2z + I112
z
,
which were obtained from (7), yields that the boundary terms are linear in
the I Kα 2 ,
⎛ τ ⎞
x −Ds Eκ (L) − Ds Eτ (L)
I2 (−κEκ (L) + τ Eτ (L) − λ(s)) + I2 I12 ⎝
y y κ ⎠
κ
E (L)
cx

Cy
⎛ ⎞
1 2 κs τ τ κ
⎜ κ Ds E(L) − κ 2 Ds E (L) + κE (L) − τ E (L)⎟
⎜ ⎟
z z z ⎜ ⎜ 1 τ
⎟
⎟
+ I2 I12 I112 ⎜ − Ds E (L) ⎟ = k.
⎜ κ ⎟
⎜ ⎟
⎝ 1 ⎠
Eτ (L)
κ

Cz
Finally, adding the products of the matrices of invariantized infinitesimals

α (I ) with the vectors C α yields the vector of invariants
⎛ ⎞
τ Eτ (L) − κEκ (L) − λ(s)
⎜ τ ⎟
⎜ −Ds Eκ (L) − Ds Eτ (L) ⎟
⎜ ⎟
⎜ κ ⎟
⎜1 κs ⎟
⎜ 2 τ τ τ κ ⎟
⎜ D E (L) − D E (L) + κE (L) − τ E (L) ⎟
⎜
υ(I ) = ⎜ κ s
κ 2 s ⎟,
⎟
⎜ τ
E (L) ⎟
⎜ ⎟
⎜ ⎟
⎜ 1 ⎟
⎜ − Ds Eτ (L) ⎟
⎝ κ ⎠
Eκ (L)
where the α (I ) are
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 0 0 0 0 0
⎜0⎟ ⎜1 0⎟ ⎜0 0 0⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜0⎟ ⎜0 0⎟ ⎜1 0 0⎟
(I ) = ⎜
x ⎟
⎜0⎟ , (I ) = ⎜
y
⎜0
⎟, (I ) = ⎜
z ⎟.
⎜ ⎟ ⎜ 0⎟⎟
⎜0
⎜ 0 κ⎟⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎝0⎠ ⎝0 0⎠ ⎝0 1 0⎠
0 0 1 0 0 0
Thus, the conservation laws are

ρF S 0
D Xρ F S Dρ F S D
⎛ ⎞
τ Eτ (L) − κEκ (L) − λ(s)
⎜ τ ⎟
⎜ −Ds Eκ (L) − Ds Eτ (L) ⎟
⎜ κ ⎟
⎜ ⎟
⎜1 2 τ κs ⎟
⎜ Ds E (L) − Ds Eτ (L) + κEτ (L) − τ Eκ (L)⎟

×⎜ ⎟ = c1 = c,
⎜κ κ 2
⎟
⎜ Eτ (L) ⎟ c2
⎜ ⎟ (44)
⎜ 1 ⎟
⎜ τ
− Ds E (L) ⎟
⎝ κ ⎠
κ
E (L)
where c1 = (c1 , c2 , c3 )T and c2 = (c4 , c5 , c6 )T are constant vectors and λ(s) is

equal to (42).
It is not surprising to see that the Frenet–Serret frame appears in this system
of conservation laws. Indeed, it is interesting to note that one is guaranteed to
find the frame from this new version of Noether’s conservation laws: as the
formulas for the laws in the original variables are known, if we write these as
far as possible in terms of κ and τ , we can then obtain Ad(ρ)−1 .
We see in the remainder of this section how the conservation laws (44) can
help reduce the integration problem.
To demonstrate this, we start by simplifying the conservation laws (44) in
two steps. These simplifications lead to an overdetermined system of equations
for x, y, and z, which is solved with relative ease once κ and τ are known as
functions of s. Finally, we give a reason behind the choice in the order in
which we solve the equations.
In the first step of the simplification, we apply an element of S E(3), say
Ad(g)−1 , to both sides of Ad(ρ(z))−1 υ(I ) = c such that it maps c1 and c2 to
the z-axis. Let Ad(g) act on c as follows:

R 0 c1 c1
Ad(g)c = = ,
D AR DRD c2 c2
where R is the three-dimensional rotation—which is the inverse of (38)

⎛ ⎞
cos β cos γ − sin α sin β cos γ − cos α sin γ − cos α sin β cos γ + sin α sin γ
⎜ ⎟
R=⎜ ⎟
⎝ cos β sin γ − sin α sin β sin γ + cos α cos γ − cos α sin β sin γ − sin α cos γ ⎠, (45)
sin β sin α cos β cos α cos β
D = diag(1, −1, 1), a diagonal matrix, and A is the matrix,

⎛ ⎞
0 −c b
⎜ ⎟
A=⎝ c 0 −a ⎠ .
−b a 0
We can easily verify that the Adjoint representation of S E(3) does not act
freely on the constant vector c, because it preserves the length of c1 and the
quantity c1T Dc2 . Indeed, to prove the latter, we multiply through
D ARc1 + DRDc1 = c2
by c1T RT D, and then obtain that
c + c1T Dc2 = c1 T D
c1 T A c2 .
1
=0
−1
Hence, let Ad(g) send c to

T
c1T Dc2
C= 0 0 |c1 | 0 0 .
|c1 |
Applying Ad(g)−1 to the conservation laws yields
Ad(g)−1 Ad(ρ(z))−1 υ(I ) = Ad(g)−1 c,
which reduces to
z))−1 υ(I ) = C
Ad(ρ( (46)
by the equivariance of the right moving frame Ad(ρ(z))−1 . Thus, it is enough
to consider

T
c1 T Dc2
C = 0 0 |c1 | 0 0 .
|c1 |
The second step of our simplification consists of applying Ad(ρ(
z)) to (46)
to obtain the following system of equations:
|c1 |
z s = υ (1) (I ), (47)
|c1 |
z
ss = υ (I ),
(2)
(48)
κ
|c1 |
xs
( yss −
ys x
ss ) = υ (I ),
(3)
(49)
κ
c1 T Dc2
|c1 |(
x
ys −
y xs ) + zs = υ (4) (I ), (50)
|c1 |
|c1 | c1 T Dc2
xss
( y−
yss
x) − z
ss = υ (I ),
(5)
(51)
κ κ|c1 |
|c1 | c1 T Dc2
( z s x
x( ss −
x
z
s ss ) −
y(
y
z
s ss −
z
y
s ss )) + xs
( yss −
ys x
ss )
κ κ|c1 |
= υ (6) (I ), (52)
where we have used υ ( j) (I ) to denote the jth component of υ(I ). Note that the
variables x,
y, and
z have become our new space coordinates after sending
c to C.
This overdetermined system of equations can be solved more easily
than (44). The two first integrals of the invariantized Euler–Lagrange
equations are obtained as follows. Define B = diag (1, 1, 1, 0, 0, 0), which
satisfies B = Ad(ρ)−T BAd(ρ)−1 . Then, we obtain the first integral of the
Euler–Lagrange equations
υ T (I )Bυ(I ) = CT BC,
τ 2
(τ Eτ (L) − κEκ (L) − λ(s))2 + −Ds Eκ (L) − Ds Eτ (L)
κ

2
1 2 τ κs
+ Ds E (L) − 2 Ds Eτ (L) + κEτ (L) − τ Eκ (L) = c12 + c22 + c32 . (53)
κ κ
This result compares to the first integral (33) for the S E(2) invariant
Lagrangians. For the second first integral, we define

0 D
D= , D = diag(1, −1, 1),
D 0
which satisfies D = Ad(ρ)−T DAd(ρ)−1 . Then, the first integral of the

Euler–Lagrange equations is
υ T (I )Dυ(I ) = CT DC,
i.e.,
τ 1
(τ Eτ (L) − κEκ (L) − λ(s)) Eτ (L) + −Ds Eκ (L) − Ds Eτ (L) Ds Eτ (L)
κ κ

1 2 τ κs
+ D E (L) − 2 Ds E (L) + κE (L) − τ E (L) Eκ (L)
τ τ κ
κ s κ
(54)
= c1 c4 − c2 c5 + c3 c6 .
We can use the above first integrals to ease the calculation of κ and τ . Once
we have solved for these, we use the system of simplified conservation laws to
solve for x,
y, and
z.
First, solving Equation (47) gives

1
z(s) = υ (1) (I ) ds. (55)
|c1 |
T
Next, multiplying Equation (49) by − c1|c1Dc
|2
2
and adding it to Equation (52) yields
|c1 | c1 T Dc2 (3)

( z s x
x( ss − x
s z
ss ) − ys z
y( ss − z
s
yss )) = υ (6) (I ) − υ (I ),
κ |c1 |2
which simplifies to

1 2 1 c1 T Dc2 (3)
|c1 | zs Ds ( x) − 1 − z
x · ss Ds (
x ·
x) = κυ (6) (I ) − κ υ (I ),
2 2 |c1 |2
where 12 Ds2 (
x ·
x) − 1 = x · x
ss and 2 Ds (
1
x ·
x) = x · xs . Setting Ds ( x · x) = h(s)
and substituting zs by (47) and z ss by its derivative yields a linear equation for h

!
Ds υ (1) (I ) c1 T Dc2 (3)
Ds h − h = 2κ υ (6)
(I ) − υ (I ) υ (1) (I ) + 2.
υ (1) (I ) |c1 |2
Solving for h, we obtain

!
1 c1 T Dc2 (3)
h(s) = υ (I )
(1)
2κ υ (I ) −
(6)
υ (I ) υ (I ) + 2 ds.
(1)
υ (1) (I ) |c1 |2
Hence,

!
1 c1 T Dc2 (3)
Ds (
x ·
x) = υ (I )
(1)
2κ υ (I ) −
(6)
υ (I ) υ (I ) +2 ds.
(1)
υ (1) (I ) |c1 |2
(56)
To solve Equations (50) and (56), we use the cylindrical coordinates
x(s) = r (s) cos θ (s), y(s) = r (s) sin θ (s), z(s) = z(s).
Starting with Equation (56), we obtain

2
r (s) = h(s) ds − z(s) ,
2
(57)
2

as Ds (
x ·
x) = Ds r (s)2 + z(s) . After applying the change of coordinates,
Equation (50) becomes

1 c1 T Dc2 (1)
r (s) θs =
2
υ (I ) −
(4)
υ (I ) ,
|c1 | |c1 |2
and then solving for θ yields

1 c1 T Dc2 (1)
θ (s) = υ (I ) −
(4)
υ (I ) ds. (58)
r (s)2 |c1 | |c1 |2
To recover x, y, and z, we apply the reverse action to
x,
y, and
z, i.e.,

x → x = R
x + a,
where R is the three-dimensional rotation (45) and a = (a, b, c)T is the
translation vector, with
⎛ ⎞
|c1 |2 cos2 β − c32
α = − tan−1 ⎝ ⎠,
c3
⎛ ⎞
c2 c3 sin β + c1 |c1 |2 cos2 β − c32
γ = tan−1 ⎝ ⎠,
c1 c3 sin β − c2 |c1 |2 cos2 β − c32
c1 c5 |c1 |2 + c2 c1 T Dc2 c2 c4 |c1 |2 − c1 c1 T Dc2

a= c+ , b= c+ ,
c3 c3 |c1 |2 c3 c3 |c1 |2
and where β and c are free.
Although only four of the equations of the system were used to solve for

x,
y, and z, we know that the remaining two equations have been satisfied.
z))−1 υ(I ) = C with respect to s and multiply
Indeed, if we differentiate Ad(ρ(
by Ad(ρ( z)), then we get
Ds υ(I ) = Ds (Ad(ρ( z))−1 υ(I ),
z))))Ad(ρ(
⎛ ⎞
0 κ 0 0 0 0
⎜−κ τ 0 ⎟
⎜ 0 0 0 ⎟
⎜ ⎟
⎜ 0 −τ 0 0 0 0 ⎟
Ds υ(I ) = ⎜
⎜ 0
⎟ υ(I ). (59)
⎜ 0 0 0 −κ 0 ⎟
⎟
⎜ ⎟
⎝ 0 0 −1 κ 0 −τ ⎠
0 −1 0 0 τ 0
The above system of equations not only forms part of an elimination ideal
of the system of equations (44), as it involves only invariants, but also
encodes the relationships among Equations (47)–(52). Thus, we see that
Ds ( Equation (47)) = κ · ( Equation (48)), and so forth. Using the Equations in
(59), we can eliminate (48) and (51) from the system.
Thus, for one-dimensional variational problems invariant under S E(3), one
needs only to solve Equations (53) and (54), which are of lower order than
the Euler–Lagrange equations in the original variables, and then one has
Equations (55), (57), and (58), which one uses to find
x = r cos θ ,
y = r sin θ ,
and z.
In the following two well-known examples, we can verify that the formulas
derived here provide the expected solutions, giving the obvious check on our
calculations.
EXAMPLE 6. For the Lagrangian with L = 1, Equations (53), (54), (55),

(57), and (58) yield the equation of a line in parametric form as the solution
that minimizes the arc length, as expected. Note that | xs | = 1 imposes a
condition on the constants of integration.

EXAMPLE 7. Consider the Lagrangian κ 2 ds with torsion, τ , set equal to
zero. Then, we obtain that κ satisfies the Euler–Lagrange equation
κss + 12 κ 3 = 0,
which is the same equation as for the S E(2) case, or the first integral
κ 4 + 4κs2 = |c1 |.
From Equation (54), we know that
c1T Dc2 = 0,
and from Equations (55), (57), and (58), we obtain

1
z(s) = − κ 2 ds,
|c1 |

2
2 1
r (s) = −
2
κ 2
ds ds − κ ds ,
2
κ2 |c1 |2
θ (s) = A,
where A is a constant. Thus, the solution
x lies on a plane that includes the

z-axis, as expected.
5. Conclusion
Noether’s First Theorem is a well-known result which provides conservation

laws for Lie group invariant variational problems. In recent work [2], the
mathematical structure of both the Euler–Lagrange system and the set of
conservation laws was given in terms of the differential invariants of the group
action and a moving frame.
In this paper, we demonstrated that the knowledge of the aforementioned
structure of Noether’s conservation laws allowed for a solution to
one-dimensional variational problems which are invariant under S E(2) and

S E(3). In summary, for both cases, once we have solved the obtained first
integrals of the invariantized Euler–Lagrange equations for the invariants,
finding the extremal curves reduces always to calculating quadratures.
References
1. E. NOETHER, Invariante variationsprobleme. Nachr. Ges. Wiss. Göttingem, Math.-Phys.

Kl. 235–257 (1918). An English translation is available at arXiv:physics/0503066v1
[physics.hist-ph].
2. T. M. N. GONÇALVES and E. L. MANSFIELD, On moving frames and Noether’s conservation
laws, Stud. Appl. Math. 128:1–29 (2012).
3. S. LIE, Klassifikation und integration von gewöhnlichen differentialgleichungen zwischen
x, y, die eine Gruppe von transformationen gestatten I, II, Mat. Ann. 32:213–281 (1888).
4. E. L. MANSFIELD, A Practical Guide to the Invariant Calculus, Cambridge University
Press, Cambridge, 2010.
5. M. FELS and P. J. OLVER, Moving coframes I, Acta Appl. Math. 51:161–312 (1998).
6. M. FELS and P. J. OLVER, Moving coframes II, Acta Appl. Math. 55:127–208 (1999).
7. I. A. KOGAN and P. J. OLVER, Invariant Euler-Lagrange equations and the invariant
variational bicomplex, Acta Appl. Math. 76:137–193 (2003).
8. E. HUBERT, AIDA Maple package: Algebraic Invariants and their Differential Algebras,
2007. Available at: http://www.inria.fr/members/Evelyne.Hubert/aida.
9. P. J. OLVER, Applications of Lie Groups to Differential Equations (2nd ed.), Springer,
New York, 1993.
10. L. EULER, Methodus inveniendi lineas curvas maximi minimive proprietate gaudentes,
sive solutio problematis isoperimetrici lattissimo sensu accepti, Opera Omnia: Series 1,
Volume 24, Lausanne, 1744.
11. R. LEVIEN, The elastica: A mathematical history, 2008. Available at: http://levien.com/
phd/elastica_hist.pdf.
UNIVERSITY OF KENT
(Received 27 June, 2012)

STUDIES IN APPLIED MATH 2013-Goncalves-Moving Frames and Conservation Laws For Euclidean Invariant Lagrangians PDF

Cargado por

Información del documento

Título original

Derechos de autor

Formatos disponibles

Compartir este documento

Compartir o incrustar documentos

Opciones para compartir

¿Le pareció útil este documento?

¿Este contenido es inapropiado?

Copyright:

Formatos disponibles

STUDIES IN APPLIED MATH 2013-Goncalves-Moving Frames and Conservation Laws For Euclidean Invariant Lagrangians PDF

Cargado por

Copyright:

Formatos disponibles

Moving Frames and Conservation Laws for Euclidean

By T. M. N. Gonçalves and E. L. Mansfield

In 1918, Emmy Noether wrote the seminal paper “Invariante Variationsprobleme”

Address for correspondence: T. M. N. Gonçalves, School of Mathematics, Statistics and Actuarial

DOI: 10.1111/j.1467-9590.2012.00566.x 134

2. Structure of Noether’s conservation laws

In this section, we give a brief overview to the theoretical foundations of

2.1. Moving frames and differential invariants of a group action

The operator ∂/∂ xi extends to the total differentiation operator

where Di maps J n into J n+1 .

Figure 1. A local foliation with a transverse cross section.

ρ(z) · z = k, {k} = O(z) ∩ K.

known as the normalization equations, for the r group parameters describing G

By the implicit function theorem, a unique solution of (1) provides an

Remark 1: As we see later, to obtain the new version of Noether’s

EXAMPLE 1. Consider the S E(2) group acting on curves in the

THEOREM 1. Let ρ(z) be a right moving frame. Then, the quantity

Consider z = (z 1 , . . . , z m ) ∈ M and let the normalization equations 

DEFINITION 1. For any prolonged action in the jet space J n (X × U ), the

EXAMPLE 1 (CONT). Evaluating 

THEOREM 2 (REPLACEMENT THEOREM [6]). If f (z) is an invariant, then

EXAMPLE 1 (CONT). We know that S E(2) preserves |xs |, thus applying

Since we are only considering variational problems with invariant independent

This leads to the following definition:

DEFINITION 2. Invariant differentiations of the jet coordinates, Ji and I Kα ,

Since ξi and φ Kα ,i are the functions of the xi , for i = 1, . . . , p, u α , for

α = 1, . . . , q, and u αK , we can define

where is the index for the group parameters and r = dim(G).

The proof of this theorem can be found on page 134 of [4].

THEOREM 4. Suppose {ψi (z) = 0, i = 1, . . . , r } is the set of normalization

Selecting the appropriate columns of the table of infinitesimals, we obtain

The transpose of the Jacobian matrix invariantized, in this case, is equal to

and the invariant total derivative matrix is

EXAMPLE 3. Suppose x = x(s, t) and y = y(s, t). Let g ∈ S E(2) act on

2.2. Invariant calculus of variations

To obtain the invariantized Euler–Lagrange equations, we first introduce

where H is an N × q matrix of operators depending only on the Di , for

Remark 2: Up to this moment, we have represented the independent

For simplicity, consider a one-dimensional Lagrangian L(x, y, yx , yx x , . . .) dx

mirroring the calculations as for the computation of the Euler–Lagrange

2.3. New version of Noether’s conservation laws

Remark 3: In Example 1, we used the right action of S E(2) on the plane

where these were obtained from (7), and rewriting (20) as

yields the boundary terms in a form that is linear in the I Kα 2 .

where α indexes a dependent variable, zi the dependent variable α and their

and the vectors Ciα = (Ci,Jα

are the infinitesimals of the prolonged group action. Let α (I ), for α = 1, . . . , q

The proof can be found in [2].

LEMMA 1. Let {I Kα } be the set of generating invariants of the prolonged

Proof : According to Definition 2, the derivative of I Kα is equal to

which by Theorems 3 and 4 is equivalent to

where (Tt∗ ) corresponds to row t of T, Equation (8). As mentioned in

after we conflate t with a group parameter a j of G, where j = 1, . . . , r and

Writing the above in matrix form yields

In the next sections, we derive formulas for the extremal curves of

3. Lagrangians invariant under SE(2)

Recall that the invariantized Euler–Lagrange equation for a one-dimensional

and its associated conservation laws are

−Ds Eκ (L) = xs c2 − ys c1 , (31)

Eκ (L) = yc1 − xc2 + c3 . (32)

EXAMPLE 4. If we consider the Lagrangian with L = 1 in Equations (33),

EXAMPLE 5. Another famous Lagrangian to consider is

where κ is the Euclidean curvature, λ the Lagrange multiplier, and η − 1 = 0

Consider z = (z 1 , . . . , z m ) ∈ M and let the normalization equations

EXAMPLE 1 (CONT). Evaluating

are the infinitesimals of the prolonged group action. Let α (I ), for α = 1, . . . , q

where the α (I ) are