Symmetry Methods for

Differential Equations
and Conservation Laws
Peter J. Olver
University of Minnesota
http://www.math.umn.edu/ ∼ olver

Santiago, November, 2010

 Symmetry Groups of
Differential Equations
System of differential equations
∆(x, u(n)) = 0
G — Lie group acting on the space of independent
and dependent variables:
(x
e, u
e ) = g · (x, u) = (Ξ(x, u), Φ(x, u))
G acts on functions u = f (x) by transforming their
graphs:

g
7−→

Definition. G is a symmetry group of the system

∆ = 0 if fe = g · f is a solution whenever f is.
 Infinitesimal Generators
Vector field:
d
v|(x,u) = g · (x, u)|ε=0
dε ε
In local coordinates:
p q
X i ∂ X ∂
v= ξ (x, u) i + ϕα(x, u) α
i=1 ∂x α=1 ∂u
generates the one-parameter group
dxi duα
= ξ i(x, u) = ϕα(x, u)
dε dε
Example. The vector field
∂ ∂
v = −u +x
∂x ∂u
generates the rotation group
x
e = x cos ε − u sin ε u
e = x sin ε + u cos ε

since
dx
e du
e
= −u
e =x
e
dε dε
 Jet Spaces

x = (x1, . . . , xp) — independent variables

u = (u1, . . . , uq ) — dependent variables

∂ k uα
uα
J = j1 — partial derivatives
∂x . . . ∂xk
(x, u(n)) = ( . . . xi . . . uα . . . uα
J ... ) ∈ J
n

— jet coordinates
!
n (n) p+n
dim J = p + q =p+q
n
 Prolongation to Jet Space
Since G acts on functions, it acts on their derivatives,
leading to the prolonged group action on the jet space:
(x e (n) ) = pr(n) g · (x, u(n))
e, u

=⇒ formulas provided by implicit differentiation

Prolonged vector field or infinitesimal generator:

X α ∂
pr v = v + ϕJ (x, u(n)) α
α,J ∂uJ
The coefficients of the prolonged vector field are given
by the explicit prolongation formula:
p
X
ϕα α
J = DJ Q + ξ i uα
J,i
i=1
Q = (Q1, . . . , Qq ) — characteristic of v
p
α (1) α
X ∂uα
i
Q (x, u )=ϕ − ξ
i=1 ∂xi
⋆ Invariant functions are solutions to
Q(x, u(1)) = 0.
 Symmetry Criterion

Theorem. (Lie) A connected group of transforma-

tions G is a symmetry group of a nondegenerate
system of differential equations ∆ = 0 if and only if
pr v(∆) = 0 (∗)
whenever u is a solution to ∆ = 0 for every infinitesi-
mal generator v of G.
(*) are the determining equations of the symmetry
group to ∆ = 0. For nondegenerate systems, this
is equivalent to
X
pr v(∆) = A · ∆ = Aν ∆ν
ν
 Nondegeneracy Conditions

Maximal Rank:
!
∂∆ν ∂∆ν
rank ··· ··· ··· = max
∂xi ∂uα
J
(n)
Local Solvability: Any each point (x0, u0 ) such that
(n)
∆(x0, u0 ) = 0
there exists a solution u = f (x) with
(n)
u0 = pr(n) f (x0)
Nondegenerate = maximal rank + locally solvable
Normal: There exists at least one non-characteristic di-
(n)
rection at (x0, u0 ) or, equivalently, there is a change
of variable making the system into Kovalevskaya form
∂ n uα α
= Γ (x, e (n) )
u
∂tn

Theorem. (Finzi) A system of q partial differential

equations ∆ = 0 in q unknowns is not normal if and
only if there is a nontrivial integrability condition:

X
D∆ = Dν ∆ν = Q order Q < order D + order ∆
ν
Under-determined: The integrability condition follows
from lower order derivatives of the equation:
e ∆≡0
D
Example:
∆1 = uxx + vxy , ∆2 = uxy + vyy
Dx∆2 − Dy ∆1 ≡ 0

Over-determined: The integrability condition is genuine.

Example:
∆1 = uxx + vxy − vy , ∆2 = uxy + vyy + uy
Dx∆2 − Dy ∆1 = uxy + vyy
 A Simple O.D.E.

uxx = 0

Infinitesimal symmetry generator:

∂ ∂
v = ξ(x, u) + ϕ(x, u)
∂x ∂u
Second prolongation:
∂ ∂
v(2) = ξ(x, u) + ϕ(x, u) +
∂x ∂u
∂ ∂
+ ϕ1(x, u(1)) + ϕ2(x, u(2)) ,
∂ux ∂uxx
 ϕ1 = ϕx + (ϕu − ξx)ux − ξuu2x,
ϕ2 = ϕxx + (2ϕxu − ξxx)ux + (ϕuu − 2ξxu)u2x −
− ξuuu3x + (ϕu − 2ξx)uxx − 3ξuuxuxx.

Symmetry criterion:
ϕ2 = 0 whenever uxx = 0.
Symmetry criterion:
ϕxx + (2ϕxu − ξxx)ux + (ϕuu − 2ξxu)u2x − ξuuu3x = 0.
Determining equations:
ϕxx = 0 2ϕxu = ξxx ϕuu = 2ξxu ξuu = 0
=⇒ Linear!
General solution:
ξ(x, u) = c1x2 + c2xu + c3x + c4u + c5
ϕ(x, u) = c1xu + c2u2 + c6x + c7u + c8
Symmetry algebra:

v1 = ∂x v5 = u∂x
v2 = ∂u v6 = u∂u
v3 = x∂x v7 = x2∂x + xu∂u
v4 = x∂u v8 = xu∂x + u2∂u

Symmetry Group:
!
ax + bu + c dx + eu + f
(x, u) 7−→ ,
hx + ju + k hx + ju + k
=⇒ projective group
 Prolongation
Infinitesimal symmetry
∂ ∂ ∂
v = ξ(x, t, u) + τ (x, t, u) + ϕ(x, t, u)
∂x ∂t ∂u
First prolongation
(1) ∂ ∂ ∂ x ∂ t ∂
pr v = ξ +τ +ϕ +ϕ +ϕ
∂x ∂t ∂u ∂ux ∂ut
Second prolongation
∂ ∂ ∂
pr(2) v = pr(1) v + ϕxx + ϕxt + ϕtt
∂uxx ∂uxt ∂utt
where
ϕx = DxQ + ξuxx + τ uxt
ϕt = DtQ + ξuxt + τ utt
ϕxx = Dx2 Q + ξuxxt + τ uxtt
Characteristic
Q = ϕ − ξux − τ ut
 ϕx = DxQ + ξuxx + τ uxt
= ϕx + (ϕu − ξx)ux − τxut − ξuu2x − τuuxut

ϕt = DtQ + ξuxt + τ utt

= ϕt − ξtux + (ϕu − τt)ut − ξuuxut − τuu2t

ϕxx = Dx2 Q + ξuxxt + τ uxtt

= ϕxx + (2φxu − ξxx)ux − τxxut
+ (φuu − 2ξxu)u2x − 2τxuuxut − ξuuu3x−
− τuuu2xut + (ϕu − 2ξx)uxx − 2τxuxt
− 3ξuuxuxx − τuutuxx − 2τuuxuxt
 Heat Equation

ut = uxx

Infinitesimal symmetry criterion

ϕt = ϕxx whenever ut = uxx
Determining equations
Coefficient Monomial
0 = −2τu uxuxt
0 = −2τx uxt
0 = −τuu u2xuxx
−ξu = −2τxu − 3ξu uxuxx
ϕu − τt = −τxx + ϕu − 2ξx uxx
0 = −ξuu u3x
0 = ϕuu − 2ξxu u2x
−ξt = 2ϕxu − ξxx ux
ϕt = ϕxx 1
General solution
ξ = c1 + c4x + 2c5t + 4c6xt
τ = c2 + 2c4t + 4c6t2
ϕ = (c3 − c5x − 2c6t − c6x2)u + α(x, t)
αt = αxx
Symmetry algebra
v1 = ∂x space transl.
v2 = ∂t time transl.
v3 = u∂u scaling
v4 = x∂x + 2t∂t scaling
v5 = 2t∂x − xu∂u Galilean
v6 = 4xt∂x + 4t2∂t − (x2 + 2t)u∂u inversions
vα = α(x, t)∂u linearity
 Potential Burgers’ equation

ut = uxx + u2x

Infinitesimal symmetry criterion

ϕt = ϕxx + 2uxϕx
Determining equations
Coefficient Monomial
0 = −2τu uxuxt
0 = −2τx uxt
−τu = −τu u2xx
−2τu = −τuu − 3τu u2xuxx
−ξu = −2τxu − 3ξu − 2τx uxuxx
ϕu − τt = −τxx + ϕu − 2ξx uxx
−τu = −τuu − 2τu u4x
−ξu = −2τxu − ξuu − 2τx − 2ξu u3x
ϕu − τt = −τxx + ϕuu − 2ξxu + 2ϕu − 2ξx u2x
−ξt = 2ϕxu − ξxx + 2ϕx ux
ϕt = ϕxx 1
General solution
ξ = c1 + c4x + 2c5t + 4c6xt
τ = c2 + 2c4t + 4c6t2
ϕ = c3 − c5x − 2c6t − c6x2 + α(x, t)e−u
αt = αxx
Symmetry algebra
v1 = ∂x
v2 = ∂t
v3 = ∂u
v4 = x∂x + 2t∂t
v5 = 2t∂x − x∂u
v6 = 4xt∂x + 4t2∂t − (x2 + 2t)∂u
vα = α(x, t)e−u∂u
Hopf-Cole w = eu maps to heat equation.
 Symmetry–Based Solution Methods
Ordinary Differential Equations
• Lie’s method
• Solvable groups
• Variational and Hamiltonian systems
• Potential symmetries
• Exponential symmetries
• Generalized symmetries
 Partial Differential Equations

• Group-invariant solutions
• Non-classical method
• Weak symmetry groups
• Clarkson-Kruskal method
• Partially invariant solutions
• Differential constraints
• Nonlocal Symmetries
• Separation of Variables
 Integration of O.D.E.’s
First order ordinary differential equation
du
= F (x, u)
dx
Symmetry Generator:
∂ ∂
v = ξ(x, u) + ϕ(x, u)
∂x ∂u
Determining equation
2 ∂F ∂F
ϕx + (ϕu − ξx)F − ξuF = ξ +ϕ
∂x ∂u
♠ Trivial symmetries
ϕ
=F
ξ
Method 1: Rectify the vector field.
v|(x0,u0) 6= 0
Introduce new coordinates
y = η(x, u) w = ζ(x, u)
near (x0, u0) so that
∂
v=
∂w
These satisfy first order p.d.e.’s
ξ ηx + ϕ ηu = 0 ξ ζx + ϕ ζ u = 1
Solution by method of characteristics:
dx du dt
= =
ξ(x, u) ϕ(x, u) 1
The equation in the new coordinates will be invariant
if and only if it has the form
dw
= h(y)
dy
and so can clearly be integrated by quadrature.
Method 2: Integrating Factor
If v = ξ ∂x + ϕ ∂u is a symmetry for
P (x, u) dx + Q(x, u) du = 0
then
1
R(x, u) =
ξP +ϕQ
is an integrating factor.
♠ If
ϕ P
=−
ξ Q
then the integratimg factor is trivial. Also, rectification
of the vector field is equivalent to solving the original
o.d.e.
 Higher Order Ordinary Differential Equations

∆(x, u(n)) = 0
If we know a one-parameter symmetry group
∂ ∂
v = ξ(x, u) + ϕ(x, u)
∂x ∂u
then we can reduce the order of the equation by 1.
Method 1: Rectify v = ∂w . Then the equation is
invariant if and only if it does not depend on w:
∆(y, w′, . . . , wn) = 0
Set v = w′ to reduce the order.
Method 2: Differential invariants.
h[ pr(n) g · (x, u(n)) ] = h(x, u(n)), g∈G
Infinitesimal criterion: pr v(h) = 0.

Proposition. If η, ζ are nth order differential invari-

ants, then
dη Dxη
=
dζ Dxζ
is an (n + 1)st order differential invariant.

Corollary. Let
y = η(x, u), w = ζ(x, u, u′)
be the independent first order differential invariants
for G. Any nth order o.d.e. admitting G as a symmetry
group can be written in terms of the differential
invariants y, w, dw/dy, . . . , dn−1w/dy n−1.

In terms of the differential invariants, the nth order

o.d.e. reduces to
f
∆(y, w(n−1)) = 0
For each solution w = g(y) of the reduced equation, we
must solve the auxiliary equation
ζ(x, u, u′) = g[η(x, u)]
to find u = f (x). This first order equation admits G as
a symmetry group and so can be integrated as before.
 Multiparameter groups

• G - r-dimensional Lie group.

Assume pr(r) G acts regularly with r dimensional
orbits.
Independent r th order differential invariants:
y = η(x, u(r)) w = ζ(x, u(r))
Independent nth order differential invariants:
dw dn−r w
y, w, , . . . , n−r .
dy dy
In terms of the differential invariants, the equation
reduces in order by r:
f
∆(y, w(n−r)) = 0
For each solution w = g(y) of the reduced equation, we
must solve the auxiliary equation
ζ(x, u(r)) = g[η(x, u(r))]
to find u = f (x). In this case there is no guarantee
that we can integrate this equation by quadrature.
Example. Projective group G = SL(2)
!
au+ b
(x, u) 7−→ x, , a d − b c = 1.
cu + d
Infinitesimal generators:
∂u, u ∂u, u2 ∂u
Differential invariants:
2 u′ u′′′ − 3 u′′2
x, w=
u′2
=⇒ Schwarzian derivative
Reduced equation
f
∆(y, w(n−3)) = 0
Let w = h(x) be a solution to reduced equation.
To recover u = f (x) we must solve the auxiliary
equation:
2 u′ u′′′ − 3 u′′2 = u′2 h(x),
which still admits the full group SL(2).
Integrate using ∂u:
u′ = z 2 z z ′′ − z ′2 = z 2 h(x)
Integrate using u ∂u = z ∂z :
v = (log z)′ 2 v ′ + v 2 = h(x)
No further symmetries, so we are stuck with a Riccati
equation to effect the solution.
 Solvable Groups

• Basis v1, . . . , vr of the symmetry algebra g such

that X k
[ vi , vj ] = cij vk , i<j
k<j
If we reduce in the correct order, then we are guaran-
teed a symmetry at each stage. Reduced equation for
subalgebra {v1, . . . , vk }:
f(k)(y, w (n−k)) = 0
∆
admits a symmetry v
e k+1 corresponding to vk+1.
Theorem. (Bianchi) If an nth order o.d.e. has a
(regular) r-parameter solvable symmetry group, then
its solutions can be found by quadrature from those of
the (n − r)th order reduced equation.
Example.
x2 u′′ = f (x u′ − u)
Symmetry group:
v = x ∂u, w = x ∂x,
[ v, w ] = − v.
Reduction with respect to v:
z = x u′ − u
Reduced equation:
x z ′ = h(z)
still invariant under w = x ∂x, and hence can be solved
by quadrature.
Wrong way reduction with respect to w:
y = u, z = z(y) = x u′
Reduced equation:
z(z ′ − 1) = h(z − y)
• No remaining symmetry; not clear how to integrate
directly.
 Group Invariant Solutions
System of partial differential equations
∆(x, u(n)) = 0
G — symmetry group
Assume G acts regularly on M with r-dimensional
orbits
Definition. u = f (x) is a G-invariant solution if
g·f =f for all g ∈ G.
i.e. the graph Γf = { (x, f (x)) } is a (locally) G-
invariant subset of M .
• Similarity solutions, travelling waves, . . .
Proposition. Let G have infinitesimal genera-
tors v1, . . . , vr with associated characteristics
Q1, . . . , Qr . A function u = f (x) is G-invariant
if and only if it is a solution to the system of first
order partial differential equations
Qν (x, u(1)) = 0, ν = 1, . . . , r.

Theorem. (Lie). If G has r-dimensional orbits,

and acts transversally to the vertical fibers
{ x = const. }, then all the G-invariant solutions
to ∆ = 0 can be found by solving a reduced sys-
tem of differential equations ∆/G = 0 in r fewer
independent variables.
Method 1: Invariant Coordinates.
The new variables are given by a complete set of
functionally independent invariants of G:
ηα(g · (x, u)) = ηα(x, u) for all g∈G
Infinitesimal criterion:
vk [ηα] = 0, k = 1, . . . , r.
New independent and dependent variables:
y1 = η1(x, u), . . . , yp−r = ηp−r (x, u)
w1 = ζ1(x, u), . . . , wq = ζ q (x, u)
Invariant functions:
w = η(y), i.e. ζ(x, u) = h[η(x, u)]
Reduced equation:
∆/G(y, w(n)) = 0
Every solution determines a G-invariant solution to
the original p.d.e.
Example. The heat equation ut = uxx
Scaling symmetry: x ∂x + 2 t ∂t + a u ∂u
x
Invariants: y=√ , w = t−au
t
u = taw(y), ut = ta−1( − 12 y w′ + a w ), uxx = taw′′.
Reduced equation
w′′ + 12yw′ − aw = 0
−y 2 /8
√
1
Solution: w=e U( 2 a + 2)
2 , y/
=⇒ parabolic cylinder function
Similarity solution:
a −x2 /8t 1
√
u(x, t) = t e U( 2 a + 2 , x/2t)
Example. The heat equation ut = uxx
Galilean symmetry: 2 t ∂x − x u ∂u
Invariants: x2 /4t
y=t w=e u
!
2 2 x2
u = e− x /4t
w(y), ut = e− x /4t ′
w + 2w ,
4t
2
!
2 x 1
uxx = e− x /4t
2
− w.
4t 2t
Reduced equation: 2 y w′ + w = 0
k 2
Source solution: w = k y −1/2, u = √ ex /4t
t
Method 2: Direct substitution:
Solve the combined system
∆(x, u(n)) = 0 Qk (x, u(1)) = 0, k = 1, . . . , r
as an overdetermined system of p.d.e.
For a one-parameter group, we solve
Q(x, u(1) ) = 0
for
∂uα ϕα p−1 X ξ i ∂uα
= n −
∂xp ξ p
i = 1 ξ ∂x
i

Rewrite in terms of derivatives with respect to x1, . . . , xp−1.

The reduced equation has xp as a parameter. Dependence on
xp can be found by by substituting back into the characteristic
condition.
 Classification of invariant solutions
Let G be the full symmetry group of the system ∆ = 0. Let
H ⊂ G be a subgroup. If u = f (x) is an H-invariant solution,
and g ∈ G is another group element, then fe = g·f is an invariant
f = g · H · g −1 .
solution for the conjugate subgroup H
• Classification of subgroups of G under conjugation.
• Classification of subalgebras of g under the adjoint action.
• Exploit symmetry of the reduced equation
 Non-Classical Method
=⇒ Bluman and Cole
Here we require not invariance of the original partial differential
equation, but rather invariance of the combined system
∆(x, u(n)) = 0 Qk (x, u(1) ) = 0, k = 1, . . . , r
• Nonlinear determining equations.
• Most solutions derived using this approach come from ordi-
nary group invariance anyway.
 Weak (Partial) Symmetry Groups
Here we require invariance of
∆(x, u(n)) = 0 Qk (x, u(1) ) = 0, k = 1, . . . , r
and all the associated integrability conditions
• Every group is a weak symmetry group.
• Every solution can be derived in this way.
• Compatibility of the combined system?
• Overdetermined systems of partial differential equations.
 The Boussinesq Equation

utt + 21 (u2)xx + uxxxx = 0

Classical symmetry group:
v 1 = ∂x v 2 = ∂t v3 = x ∂x + 2 t ∂t − 2 u ∂u
For the scaling group
− Q = x ux + 2 t u t + 2 u = 0
Invariants:
!
x 1 x
y=√ w = tu u= w √
t t t
Reduced equation:
w′′′′ + 12 (w2)′′ + 41 y 2w′′ + 74 y w′ + 2w = 0
 utt + 21 (u2)xx + uxxxx = 0
Group classification:
v 1 = ∂x v 2 = ∂t v3 = x ∂x + 2 t ∂t − 2 u ∂u
Note:
Ad(ε v3) v1 = eε v1 Ad(ε v3)v2 = e2 ε v2
Ad(δ v1 + ε v2)v3 = v3 − δ v1 − ε v2
so the one-dimensional subalgebras are classified by:
{ v3 } { v1 } { v2 } { v1 + v2 } { v1 − v2 }
and we only need to determine solutions invariant under
these particular subgroups to find the most general group-
invariant solution.
 utt + 12 (u2)xx + uxxxx = 0
Non-classical: Galilean group
v = t ∂x + ∂t − 2 t ∂u
Not a symmetry, but the combined system
utt + 12 (u2)xx + uxxxx = 0 − Q = t ux + ut + 2 t = 0
does admit v as a symmetry. Invariants:
y = x − 12 t2, w = u + t2 , u(x, t) = w(y) − t2
Reduced equation:
w′′′′ + ww′′ + (w′)2 − w′ + 2 = 0
 utt + 12 (u2)xx + uxxxx = 0
Weak Symmetry: Scaling group: x ∂x + t ∂t
Not a symmetry of the combined system
utt + 21 (u2)xx + uxxxx = 0 Q = x ux + t u t = 0
x
Invariants: y = u Invariant solution: u(x, t) = w(y)
t
The Boussinesq equation reduces to
t−4w′′′′ + t−2 [(w + 1 − y)w′′ + (w′)2 − y w′] = 0
so we obtain an overdetermined system
w′′′′ = 0 (w + 1 − y)w′′ + (w′ )2 − y w′ = 0
Solutions: w(y) = 32 y 2 − 1, or w = constant
2 x2
Similarity solution: u(x, t) = 2 − 1
3t
 Symmetries and
Conservation Laws
 Variational problems
Z
L[ u ] = L(x, u(n)) dx
Ω
Euler-Lagrange equations
∆ = E(L) = 0
Euler operator (variational derivative)
α δL X ∂L
E (L) = α = (−D)J α
δu J ∂uJ

Theorem. (Null Lagrangians)

E(L) ≡ 0 if and only if L = Div P
Theorem. The system ∆ = 0 is the Euler-Lagrange
equations for some variational problem if and only if
the Fréchet derivative D∆ is self-adjoint:
∆D∗ = D .
∆

=⇒ Helmholtz conditions
 Fréchet derivative
Given P (x, u(n)), its Fréchet derivative or formal
linearization is the differential operator DP defined
by
d
DP [ w ] = P [ u + εw ]
dε ε=0
Example.
P = uxxx + uux
DP = Dx3 + uDx + ux
Adjoint (formal)
X J ∗ X
D= AJ D D = (−D)J · AJ
J J

Integration by parts formula:

P DQ = Q D∗P + Div A
where A depends on P, Q.
 Conservation Laws

Definition. A conservation law of a system of partial

differential equations is a divergence expression
Div P = 0
which vanishes on all solutions to the system.
P = (P1(x, u(k)), . . . , Pp(x, u(k)))
=⇒ The integral Z
P · dS
is path (surface) independent.
If one of the coordinates is time, a conservation law
takes the form
DtT + Div X = 0
T — conserved density X — flux
By the divergence theorem,
Z b Z bZ
(k)
T (x, t, u ) dx) = X · dS dt
Ω t=a a Ω
depends only on the boundary behavior of the solution.
R
• If the flux X vanishes on ∂Ω, then Ω T dx is
conserved (constant).
 Trivial Conservation Laws
Type I If P = 0 for all solutions to ∆ = 0, then
Div P = 0 on solutions too
Type II (Null divergences) If Div P = 0 for all
functions u = f (x), then it trivially vanishes on
solutions.
Examples:
Dx(uy ) + Dy (−ux) ≡ 0
∂(u, v) ∂(u, v) ∂(u, v)
Dx + Dy + Dz ≡0
∂(y, z) ∂(z, x) ∂(x, y)
Theorem.
Div P (x, u(k)) ≡ 0
for all u if and only if
P = Curl Q(x, u(k))
i.e. p
X
Pi = Dj Qij Qij = − Qji
j =1
Two conservation laws P and Pe are equivalent if they
differ by a sum of trivial conservation laws:
P = Pe + PI + PII
where
PI = 0 on solutions Div PII ≡ 0.
Proposition. Every conservation law of a system
of partial differential equations is equivalent to a
conservation law in characteristic form
X
Div P = Q · ∆ = Qν ∆ν
ν

Proof : X
Div P = QJν DJ ∆ν
ν,J
Integrate by parts:
X X
Div Pe = (−D) J
QJν · ∆ν Qν = (−D)J QJν
ν,J J
Q is called the characteristic of the conservation law.
Theorem. Q is the characteristic of a conservation
law for ∆ = 0 if and only if
D∗ Q + D∗ ∆ = 0.
∆ Q

Proof :
∗ Q + D∗ ∆
0 = E(Div P ) = E(Q · ∆) = D∆ Q
 Normal Systems
A characteristic is trivial if it vanishes on solutions.
Two characteristics are equivalent if they differ by a
trivial one.

Theorem. Let ∆ = 0 be a normal system of

partial differential equations. Then there is a one-to-
one correspondence between (equivalence classes of)
nontrivial conservation laws and (equivalence classes
of) nontrivial characteristics.
 Variational Symmetries

Definition. A (restricted) variational symmetry is

a transformation (x
e, u
e ) = g · (x, u) which leaves the
variational problem invariant:
Z Z
L(x e (n) ) dx
e, u e = L(x, u(n)) dx
e
Ω Ω
Infinitesimal criterion:
pr v(L) + L Div ξ = 0

Theorem. If v is a variational symmetry, then it is a

symmetry of the Euler-Lagrange equations.
⋆ ⋆ But not conversely!
Noether’s Theorem (Weak version). If v generates
a one-parameter group of variational symmetries of a
variational problem, then the characteristic Q of v is
the characteristic of a conservation law of the Euler-
Lagrange equations:
Div P = Q E(L)
 Elastostatics
Z
W (x,∇u) dx — stored energy
x, u ∈ Rp, p = 2, 3

Frame indifference
u 7−→ R u + a, R ∈ SO(p)

Conservation laws = path independent integrals:

Div P = 0.
1. Translation invariance
∂W
Pi =
∂uα
i
=⇒ Euler-Lagrange equations
2. Rotational invariance
∂W ∂W
P i = uα
i − uβi
∂uβj ∂uα
j

3. Homogeneity : W = W (∇u) x 7−→ x + a

p
X ∂W
Pi = uα
j
α − δ i
j W
α=1 ∂u i
=⇒ Energy-momentum tensor
4. Isotropy : W (∇u · Q) = W (∇u) Q ∈ SO(p)
p
k α ∂W
X
Pi = (x uk − x uj ) α + (δji xk − δki xj )W
j α
α=1 ∂ui
5. Dilation invariance : W (λ∇u) = λnW (∇u)
p
n−p X ∂W
Pi = (uαδji − xj uα
j ) α + x i
W
n α,j = 1 ∂ui
5A. Divergence identity
Div Pe = p W
p
X ∂W
Pe i = (uαδji − xj uα
j) α + xi W
j =1 ∂ui
=⇒ Knops/Stuart, Pohozaev, Pucci/Serrin
 Generalized Vector Fields
Allow the coefficients of the infinitesimal generator to
depend on derivatives of u:
p q
X ∂ X ∂
v= ξ i(x, u(k)) i + ϕα(x, u(k)) α
i=1 ∂x α=1 ∂u
Characteristic :
p
X
Qα(x, u(k)) = ϕα − ξ i uα
i
i=1

Evolutionary vector field:

q
X (k) ∂
vQ = Qα(x, u ) α
α=1 ∂u
Prolongation formula:
p
X
pr v = pr vQ + ξ iDi
i=1
X J ∂ X ∂
pr vQ = D Qα α Di = uα
J,i
α,J ∂uJ α,J ∂uαJ
=⇒ total derivative
 Generalized Flows
• The one-parameter group generated by an evolu-
tionary vector field is found by solving the Cauchy
problem for an associated system of evolution
equations
∂uα
= Qα(x, u(n)) u|ε=0 = f (x)
∂ε
 ∂
Example. v = generates the one-parameter
∂x
group of translations:
(x, y, u) 7−→ (x + ε, y, u)
Evolutionary form:
∂
vQ = −ux
∂x
Corresponding group:
∂u
= −ux
∂ε
Solution
u = f (x, y) 7−→ u = f (x − ε, y)
 Generalized Symmetries
of Differential Equations

Determining equations :
pr v(∆) = 0 whenever ∆=0
For totally nondegenerate systems, this is equivalent to
X
pr v(∆) = D∆ = Dν ∆ν
ν

⋆ v is a generalized symmetry if and only if its

evolutionary form vQ is.
• A generalized symmetry is trivial if its characteristic
vanishes on solutions to ∆. Two symmetries are equivalent
if their evolutionary forms differ by a trivial symmetry.
 General Variational Symmetries

Definition. A generalized vector field is a variational

symmetry if it leaves the variational problem invariant
up to a divergence:
pr v(L) + L Div ξ = Div B
⋆ v is a variational symmetry if and only if its evolu-
tionary form vQ is.
f
pr vQ(L) = Div B
Theorem. If v is a variational symmetry, then it is a
symmetry of the Euler-Lagrange equations.
Proof :
First, vQ is a variational symmetry if
pr vQ(L) = Div P.
Secondly, integration by parts shows
pr v (L) = D (Q) = QD∗ (1)+Div A = QE(L)+Div A
Q L L
for some A depending on Q, L. Therefore
∗ Q + D∗ ∆
0 = E(pr vQ(L)) = E(QE(L)) = E(Q ∆) = D∆ Q
∗ ∆ = pr v (∆) + D∗ ∆
= D∆Q + DQ Q Q
Noether’s Theorem. Let ∆ = 0 be a normal system
of Euler-Lagrange equations. Then there is a one-to-
one correspondence between (equivalence classes of)
nontrivial conservation laws and (equivalence classes
of) nontrivial variational symmetries. The characteris-
tic of the conservation law is the characteristic of the
associated symmetry.
Proof : Nother’s Identity:
QE(L) = pr vQ(L) − Div A = Div(P − A)
 The Kepler Problem

µx
x+ L = 12 x2 − µr
 

3
=0
r
Generalized symmetries:
v = (x · x)∂x + x(x · ∂x) − 2 x(x · ∂x)
  

Conservation law
pr v(L) = DtR
where
µx
R = x ∧ (x ∧ x) −
 

r
=⇒ Runge-Lenz vector
Noether’s Second Theorem. A system of Euler-
Lagrange equations is under-determined if and only if
it admits an infinite dimensional variational symmetry
group depending on an arbitrary function. The associ-
ated conservation laws are trivial.
Proof : If f (x) is any function,
f (x)D(∆) = ∆ D∗(f ) + Div P [f, ∆].
Set
Q = D∗(f ).
Example. Z Z
(ux + vy )2 dx dy
Euler-Lagrange equations:
∆1 = E u(L) = uxx + vxy = 0

∆2 = E v (L) = uxy + vyy = 0

Dx∆2 − Dy ∆2 ≡ 0
Symmetries
(u, v) 7−→ (u + ϕy , v − ϕx)