The Diffiety Institute Preprint Series

Preprint DIPS–2/98 January 26, 1998

Symmetries and recursion operators

for soliton equations

by

I.S. Krasil’shchik

Available via INTERNET:

http://ecfor.rssi.ru/˜diffiety/
http://www.botik.ru/˜diffiety/
anonymous FTP:
ftp://ecfor.rssi.ru/general/pub/diffiety
ftp://www.botik.ru/˜diffiety/preprints

The Diffiety Institute

Polevaya 6-45, Pereslavl-Zalessky, 152140 Russia.
 Symmetries and recursion operators for
soliton equations
I.S. Krasil’shchik∗
Moscow Institute for Municipal Economy
and the Diffiety Institute †
June 24, 1998

Abstract
The theory of symmetries for nonlinear differential equations is exposed
using naı̈ve coordinate language. As an example, the Burgers equation
is considered. Methods for computing recursion operators are given and
nonlocal problems arising in this context are discussed.
Key words and phrases: nonlinear differential equations, symmetries,
recursion operators, coverings
1991 AMS Subject Classification: 58F37.

It is known that all soliton equations possess several structures which make of
them very interesting objects for mathematical investigations. These structures
are:

• infinite series of commuting symmetries and conservation laws,

• Hamiltonian (and usually, bi-Hamiltonian) maps,

• recursion operators.

The first equation which was found to possess all these properties was the cele-
brated Korteweg-de Vries equation

ut = uxxx + uux . (1)

∗
This work was partially supported by the Russian Foundation for Base Research Grant N
97-01-00462 and by the INTAS Grant N 96-0793.
†
For correspondence: 1st Tverskoy-Yamskoy per. 14, Apt.45, 125047 Moscow, Russia. E.mail:
[email protected]

1
It was shown, in particular, that soliton solutions of (1) were invariant (with
respect to higher symmetries) solutions. Later similar results were obtained for
other soliton equations.
In fact, geometrical theory of differential equations1 originates from classical
works by S. Lie and A. Bäcklund and much later was revised and put into the
modern differential geometry framework in [3, 5, 6].
In this paper we do not intend to expose the general theory of symmetries for
differential equations, but try to clarify practical aspects of computations. As an
example, we choose the Burgers equation

ut = uxx + uux (2)

for which the whole conceptual scheme looks quite transparent. Besides symme-
tries, we explain how to compute recursion operators using the methods developed
in [1, 2]. Recursion operators are a very powerful instrument for symmetry com-
putation, but as a rule need the use of nonlocal variables. Hence, we discuss the
latter ones as well.

1 Symmetries of differential equations.

Let us analyse the notion of a symmetry and deduce efficient formulae for com-
putations.
What is a symmetry? Consider a differential equation2 E given by

∂ |σ|u
F (x, u, . . . , , . . .) = 0, (3)
∂xσ
where x = (x1, . . . , xn) are independent variables, u = (u1 , . . . , um ) are unknown
|σ| |σ|
functions (dependent variables) and ∂∂xσu = i∂1 u in for σ = (i1, . . . , in), |σ| =
∂x1 ...∂xn
i1 + . . . + in ≤ k, k being the order of E.
Let
∂u ∂ |σ|u
= Φ(x, u, . . . , , . . .) (4)
∂τ ∂xσ
be an evolution equation and suppose (3) and (4) to be compatible. Let u = u(x)
be a solution of (3) and uτ = u(x, τ ) be a solution of (4) and (3) with initial data
u0 = u. Then {uτ } is a curve in the space of solutions of equation (3). Taking
an arbitrary solution3 of (3), we get a correspondence Φτ : u 7→ uτ . Hence,
evolution equation (4) determines a one-parameter family of transformations on
the solution space of (3). Thus we can give a preliminary
1
Of which the theory of symmetries is a natural component.
2
When we say a differential equation we in fact mean a system of equations and thus F in
(3) is a vector function F = (F 1 , . . . , F r ).
3
We assume the initial value problem for (4) to be uniquely solvable.

2
Definition 1 Symmetry is an evolution equation which preserves solutions.

Of course, this is not an operational definition: the concept of a solution space,

as well as of an action on this space, remains rather vague. To make them exact,
we should answer the next question:
What is a differential equation? Groups of transformations arise in differ-
ential geometry as maps of main objects (curves, surfaces, etc.) preserving basic
structures (e.g., Riemannian metrics). Thus, if one is going to use differential ge-
ometry techniques in application to differential equations, one should put these
equations into appropriate conceptual framework. This is being done as follows.
Consider the space J k (n, m) with coordinates

(x1, . . . , xn , u1 , . . . , um , . . . , u1σ , . . . , um
σ , . . .),

where, as above, σ = (i1, . . . , in ) and |σ| ≤ k. This space is called the space of
jets of the order k. Then an equation E of the form (3) can be understood as a
subset in J k (n, m) defined by relations

F 1(x, u, . . . , uσ , . . .) = 0, . . . , F r (x, u, . . . , uσ , . . .) = 0.

If all F i are smooth functions and the system F 1, . . . , F r is of the maximal rank,
then this subset is a smooth4 submanifold of the codimension r. Thus, differential
equations are smooth submanifolds in jet spaces.
Let f = (f 1 (x), . . . , f m (x)) be smooth vector function. Then f determines an
n-dimensional submanifold in J k (n, m) defined by

∂ |σ| f j
ujσ = , j = 1, . . . , m, |σ| ≤ k,
∂xσ
(here and below we formally assume u(0,...,0) = u). This submanifold is called the
k-jet of f and is denoted by jk (f). It easily seen that f satisfies equation (3) if
and only if jk (f) lies in corresponding submanifold in J k (n, m). In other words,
solutions of E are identified with jets lying in the equation manifold. Hence, the
following definition is justified:

Definition 2 Let E ⊂ J k (n, m) be a differential equation. A symmetry of E is

a diffeomorphism ϕ : J k (n, m) → J k (n, m), such that (i) ϕ(E) = E and (ii)
ϕ(jk (f)) is of the form jk (fϕ ) for any smooth function f.

This definition, being perfectly correct, can not be considered as a working one
though: to compute symmetries one needs to know solutions, and it does not
make sense!
4
By smooth we always mean of class C ∞ .

3
Infinite prolongations and the Cartan distribution. To overcome this prob-
lem, first note the following. Let f(x) be a solution of (3). Differentiate F j (f(x))
along xα :

∂ F j (f(x))
=
∂xα
∂F j ∂F j ∂f1 ∂F j ∂fm
= (f(x)) + (f(x)) + . . . + (f(x)) +...
∂xα ∂u1 ∂xα ∂um ∂xα
∂F j ∂ |σ|+1 f 1 ∂F j ∂ |σ|+1f m
...+ (f(x)) + . . . + (f(x)) +...
∂u1σ ∂xσ+1α ∂um σ ∂xσ+1α
def
where σ + 1α = (i1, . . . , iα + 1, . . . , in ). Define operators

∂ X
m X
∂
Dα[k] = + ujσ+1α j (5)
∂xα j=1 |σ|≤k ∂uσ

Then one can easily see that the system

F = 0, D1[k] (F ) = 0, . . . , Dn[k] (F ) = 0 (6)

possesses the same solutions as the initial equation (3). If E ⊂ J k (n, m) is the
submanifold corresponding to (3), then a submanifold in J k+1 (n, m) corresponds
to (6), which we denote by E 1 and call the first prolongation of E.
Now, we define by induction E l+1 = (E l )1 and see that the set E l lies in
J k+l (n, m); we call E l the l-th prolongation of E. It is reasonable to consider all
prolongations of E simultaneously. What geometrical object does correspond to
this construction?
Let J ∞(n, m) be the space with coordinates

(x1, . . . , xn , u1 , . . . , um , . . . , u1σ , . . . , um
σ , . . .),

where |σ| is unlimited. This space is called the space of infinite jets and the
whole series of prolongations of the equation E determines a submanifold E ∞
in J ∞ (n, m) which is called the infinite prolongation of the equation E. Infinite
prolongations are expressed by relations

Dσ (F ) = 0, |σ| ≥ 0,

where Dσ = D1i1 ◦ . . . ◦ Dnin and

∂ X j ∂
Dα = + uσ+1α j (7)
∂xα j,σ ∂uσ

are infinite counterparts of (5) and are called total derivatives.

4
 Expressions of the form (7) may be viewed at as vector fields on the space
J ∞ (n, m). Thus, at any point θ ∈ J ∞ (n, m) an n-dimensional plane Cθ arises
spanned by vectors of these fields. It means that there exists a distribution C :
θ 7→ Cθ on J ∞ (n, m). We call it the Cartan distribution. Its importance to the
theory of differential equations is explained by the following

Theorem 1 [3] (i) If E ∞ ⊂ J ∞ (n, m) is the infinite prolongation, then Cθ is

tangent to E ∞ at any point θ ∈ E ∞ .
A submanifold in E ∞ is a solution of E if and only if it is an integrable
manifold of the Cartan distribution.

Hence, E ∞ carries the structure which completely determines solutions of E, and

this structure is the Cartan distribution.

Definition 3 (final) A diffeomorphism ϕ : E ∞ → E ∞ is called a (higher) sym-

metry of E, if it preserves the Cartan distribution, i.e. if ϕ∗ (Cθ ) = Cϕ(θ) for any
θ ∈ E ∞.

This definition is in a complete agreement with the general geometrical approach.

The only problem consists in efficient computation of symmetries. And in fact,
everything becomes efficient, if one chooses
Infinitesimal approach. Let M be a smooth manifold and D be distribution
on M. Suppose that D is spanned by vector fields X1 , . . . , Xn . Assume that
a one-parameter group of symmetries {Aτ } of the distribution D is given, i.e.
(Aτ )∗ Dθ = DAτ (θ) for any point θ ∈ M. It is equivalent to the following system
of equations

(Aτ )∗ (Xα ) = λ1α X1 + . . . + λnα Xn , α = 1, . . . , n, (8)

where λβα are depending on τ smooth functions on M with det kλβα k 6= 0. Let X
be the vector field corresponding to {Aτ }, i.e. dA τ
dτ τ =0
= X. Differentiating (8)
with respect to τ at τ = 0 one obtains

[X, Xα ] = µ1α X1 + . . . + µnα Xn , α = 1, . . . , n, (9)

where µβα are smooth functions on M.

Definition 4 A vector field X satisfying (9) is called an infinitesimal symmetry

of the distribution D.

Symmetries of D form a Lie algebra with respect to commutator of vector fields:

if X and Y satisfy (9), then [X, Y ] satisfies it as well. Denote this algebra by
s(D).
Note now that any vector field X lying in D is a symmetry of D and, by
definition, is tangent to any integral manifold of this distribution. It means, if

5
Aτ is a one-parameter group corresponding to X, then any transformation Aτ
moves an integrable manifold of D along itself and thus trivially acts on the
space of integral manifolds as a whole. In this sense, such symmetries are trivial.
Respectively, field lying in D are trivial infinitesimal symmetries and we denote
the set of these symmetries by s0 (D).
Obviously, s0 (D) is an ideal in the Lie algebra s(D): if X ∈ s(D) and Y ∈
s0 (D), then [X, Y ] ∈ s0 (D). It means that one can consider the quotient Lie
algebra
def
sym(D) = s(D)/s0 (D),
and its elements are naturally identified with nontrivial symmetries of D.
We apply now this construction to the Cartan distribution C.
How to compute infinitesimal symmetries? Consider the space of infinite
jets J ∞ (n, m) and the infinite prolongation E ∞ of equation E ⊂ J k (n, m). Let C
be the Cartan distribution on J ∞ (n, m) and CE be its restriction onto E ∞ .

Definition 5 Elements of sym(C) are called infinitesimal symmetries of the infi-

nite jet space and their set is denoted by symn,m . Elements of sym(CE ) are called
infinitesimal symmetries of the equation E and their set is denoted by sym(E).

Both sym(CE ) and sym(E) are Lie algebras.

In the sequel we omit the adjective ”infinitesimal”, since from now on we work
with infinitesimal symmetries only.
In principal, there exist two approaches to computation of symmetries of
equation E. The first, exterior, one consists of finding those symmetries X ∈
symn,m which are tangent to E ∞ . The second, interior, is the direct computation of
sym(E). While the former seems to be preferable from computational viewpoint,
the latter is, at first glance, more justified. But, as it happens, these approaches
are equivalent.

Proposition 1 [3] Any symmetry X ∈ sym(E) is a restriction onto E of some

symmetry X 0 ∈ symn,m .

Thus we can start with description of symn,m .

First note the following. Any vector field on J ∞ (m, n) is represented
X ∂ X ∂
X= Aα + Bσβ β , (10)
α ∂xα β,σ ∂uσ

where Aα, Bσβ are smooth functions on J ∞(n, m). We say that the field X is
vertical, if it vanishes on functions depending on x only. Hence, vertical fields
P
are of the form β,σ Bσβ ∂u∂ β . Consider an arbitrary field (10). Then the field X −
P σ

α X α D α , where D α , α = 1, . . . , n are total derivatives (7), is a vertical field.

6
Hence, the algebra D(J ∞ (n, m)) of all vector fields on J ∞ (n, m) splits into the
direct sum
D(J ∞ (n, m)) = Dv (J ∞(n, m)) ⊕ s0 (C)
Dv (J ∞ (n, m)) being the algebra of vertical fields. This presentation induces a
splitting of the algebra s(C) and we obtain
s(C) = sym ⊕s0(C). (11)
n,m

From the latter it easily follows

Proposition 2 Elements of symn,m are identified with vertical vector fields X
such that
[X, Dα ] = 0 (12)
for all α = 1, . . . , n.
Using this result, let us obtain an explicit description of symmetries on J ∞ (n, m).
P
Let X = β,σ Bσβ ∂u∂ β be a vertical field. Then
σ
 
X ∂ ∂ X β ∂
[X, Dα ] =  Bβ uσ+1α β  =
σ β, +
β,σ ∂uσ ∂xα β,σ ∂uσ
X β ∂ X ∂
Bσ+1α β − Dα (Bσβ ) β ,
β,σ ∂uσ β,σ ∂uσ
β
from where one has Bσ+1 α
= Dα (Bσβ ), or
β
Bσβ = Dσ (B(0,...,0) ).
def
β
Let B β = B(0,...,0) .
Theorem 2 [3] Any symmetry X ∈ symn,m is of the form
X ∂
B = Dσ (B j ) , (13)
β,σ ∂uβσ
where B = (B 1, . . . , B m ), B β being arbitrary functions on J ∞ (n, m).
Definition 6 Vector fields B are called evolution fields, B being called the gen-
erating function.

Remark 1 Suppose, a field B is integrable5, i.e. possesses a trajectory in

J ∞ (n, m). Then evolution of uβσ along these trajectories would be governed by
equations
∂ uσ
= Dσ (B)
∂τ
5
Contrary to the finite dimensional case, this is not usually true for infinite dimensional
manifolds J ∞ (n, m) or E ∞ .

7
and, in particular,
∂u
= B(x, . . . , uσ , . . .),
∂τ
which explains the name ”evolution” for fields B and is in complete agreement
with the naı̈ve Definition 1!
Recall that symn,m is a Lie algebra, i.e. is closed under commutator of vector
fields. Hence, for any B1 and B2 their commutator [B2 , B2 ] is again an evo-
lution field B for some new B. This function B is called the Jacobi bracket of
B1 and B2 and is denoted by {B1, B2 }. From (13) it follows that
X j!
∂ Bj β ∂ B1
j
B = Dσ (B1β ) β2 − Dσ (B2 ) β . (14)
β,σ ∂uσ ∂uσ
Clearly, the function algebra on J ∞ (n, m) is a Lie algebra with respect to the
Jacobi bracket.
Now we restrict everything above said onto E ∞ . Suppose, as in (3), E is given
by relations F 1 = 0, . . . , F r = 0. Then a field X ∈ D(J ∞ (n, m)) is tangent to E ∞
if and only if one has X
X(F j ) = ajβ β
σ Dσ (F )
β,σ

for some functions ajβ

σ . In particular, if X = B , the last equation rewrites as
X
j
B (F )= ajβ β
σ Dσ (F ),
β,σ

or, if we use the notation

def
`F (B) = B (F ) (15)
as X
`F j (B) = ajβ β
σ Dσ (F ). (16)
β,σ

Definition 7 The operator

X ∂F
`jF = Dσ , j = 1, . . . , m, (17)
β ∂ujσ
is called the operator of universal linearization.
Thus, B is a symmetry of E if and only if B satisfies equation (16).
Restrict (16) onto E ∞ . Then the right-hand side vanishes by definition, and
we obtain the following important result.
Theorem 3 [3] Let E ⊂ J k (n, m)be a differential equation given by relations
F 1 = 0, . . . , F r = 0 and B be an evolution vector field. Denote by B̄ restriction
of B onto E ∞ . Then B is a symmetry of E if and only if
`E (B̄) = 0, (18)

8
where `E is the restriction of `F = (`F 1 , . . . , `F r ) onto E ∞ . In other words,

sym(E) = ker(`E ).

Note that solutions of (18) are closed under the Jacobi bracket restricted onto
E ∞.
Equation (18) is the basis for direct computations of symmetries. We illustrate
its use in the next section.

2 Example: the Burgers equation.

Consider the Burgers equation B

ut = uxx + uux (19)

and choose internal coordinates on B ∞ by setting uk = u(k,0). Below we use the

method described in [3].
Defining equations. Rewrite restrictions onto B ∞ of all basic concepts in these
coordinate system.
For the total derivatives we obviously obtain
∞
X
∂ ∂
Dx = + ui+1 , (20)
∂x k=0 ∂ui
∞
X
∂ ∂
Dt = + Dxi (u3 + u0u1 ) . (21)
∂t k=0 ∂ui

The operator of universal linearization for B is then looks as

`B = Dt − u1 − u0Dx − Dx2 (22)

and, as it follows from Theorem 3, evolution field

∞
X ∂
ϕ = Dxi (ϕ) , (23)
i=1 ∂ui

is a symmetry for B if and only if ϕ = ϕ(x, t, u0, . . . , uk ) satisfies the equation

Dt ϕ = u1ϕ + u0 Dx ϕ + Dx2 ϕ, (24)

where Dt , Dx are given by (20), (21). Computing Dx2 ϕ we obtain

∂ 2ϕ Xk
∂ 2ϕ Xk
∂ 2ϕ X
k
∂ϕ
Dx2 ϕ = 2
+ 2 u i+1 + u i+1 u j+1 + ui+2 ,
∂x i=1 ∂x∂ui i,j=0 ∂ui∂uj i=0 ∂ui

9
while !
X
i
i
Dxi (u0u1 + u3) = uαui−α+1 + ui+3 .
α=0 α
Hence, (24) transforms to
!
∂ϕ Xk X i
i ∂ϕ
+ uαui−α+1 =
∂t i=1 α=1 α ∂ui
∂ ϕ ∂ 2ϕ Xk
∂ 2ϕ Xk
∂ 2ϕ
u1 ϕ + u0 + 2 +2 ui+1 + ui+1 uj+1 . (25)
∂x ∂x i=1 ∂x∂ui i,j=0 ∂ui∂uj

Higher order terms. Note now that the left-hand side of (25) is independent
of uk+1 while the right-hand one is quadratic in this variable and is of the form
!
∂ 2ϕ ∂ 2ϕ X
k−1
∂ 2ϕ
u2k+1 2 + 2uk+1 + ui+1 .
∂uk ∂x∂uk i=0 ∂ui ∂uk
It means that
ϕ = Auk + ψ, (26)
where A = A(t) and ψ = ψ(t, x, u0, . . . , uk−1 ). Substituting (26) into equation
(25) one obtains
! !
XX
∂ ψ k−1 i
i ∂ψ X k
k
Ȧuk + + uαui−α+1 + ui uk−i+1 A =
∂t i=1 α=1 α ∂ui i=1 i
∂ ψ ∂ 2ψ X
k−1
∂ 2ψ
u1(Auk + ψ) + u0 + + 2 u i+1 +
∂x ∂x2 i=1 ∂x∂ui
X
k−1
∂ 2ψ
ui+1 uj+1 ,
i,j=0 ∂ui ∂uj
def
where Ȧ = dA dt
. Here again everything is at most quadratic in uk , and equating
coefficients at u2k and uk we get
∂ 2ψ
= 0,
∂u2k−1
!
X
k−2
∂ 2ψ ∂ 2ψ
2 ui+1 + = ku1A + Ȧ.
i=0 ∂ui ∂uk−1 ∂x∂uk−1
Hence,
1
ψ = (ku0 A + Ȧx + ȧ)uk−1 + O[k − 2],
2
where a = a(t) and O[l] denotes a function independent of ui , i > l. Thus
1
ϕ = Auk + (ku0 A + Ȧx + ȧ)uk−1 + O[k − 2] (27)
2

10
which gives the ”upper estimation” for solutions of (24).
Estimating Jacobi brackets. Let

ϕ = ϕ(t, x, u0, . . . , uk ), ψ(t, x, u0, . . . , ul )

be two symmetries of B. Then their Jacobi bracket restricted onto B ∞ looks as

X
l
∂ψ X k
∂ϕ
{ϕ, ψ} = Dxi (ϕ) − Dxj (ψ) . (28)
i=0 ∂ui i=0 ∂uj

Suppose ϕ is of the form (27) and similarly

1
ψ = Bul + (lu0B + Ḃx + ḃ)ul−1 + O[l − 2]
2
and compute (28) for these functions temporary denoting ku0 A + Ȧ + a and
lu0B + Ḃ + b by Ā and B̄ respectively. Then we have:
1 1 1
{ϕ, ψ} = Dxl (Auk + Āuk−1 )B + Dxl−1 (Auk + Āuk−1 )B̄ −
2 2 2
1 1 1
Dxk (Bul + B̄ul−1 )A − Dxk−1 (Buk + B̄ul−1)Ā + O[k + l − 1] =
2 2 2
1 1 1
(lDx (Ā)uk+l−2 + Āuk+l−1 )B̄ + (Auk+l−1 + Āuk+l−2 )B −
2 2 2
1 1 1
(kDx (B̄)uk+l−2 + B̄uk+l−1 )Ā − (Buk+l−1 + B̄uk+l−2 )A +
2 2 2
O[k + l − 3],

or in short,
1
{ϕ, ψ} = (lȦB − K ḂA)uk+l−2 + O[k + l − 3]. (29)
2
Low-order symmetries. Take k = 2 and solve equation (24) directly. Then one
obtains five independent solutions which are

ϕ01 = u1,
ϕ11 = tu1 + 1,
ϕ02 = u2 + u0 u1,
1 1
ϕ12 = tu2 + (tu0 + x)u1 + u0,
2 2
ϕ22 = t2u2 + (t2u0 + tx)u1 + tu0 + x. (30)

Action of low-order symmetries. Let us compute the action

def
Tij = {ϕji , •} = ϕj − `ϕj
i i

11
of symmetries ϕji on other symmetries of the equation B.
For ϕ01 one has
X ∂ ∂
T10 = u1 − `u1 = ui+1 − Dx = − .
i≥0 ∂ui ∂x

Hence, if ϕ = Auk + O[k − 1] is a function of the form (27), then one has
1
T10ϕ = − Ȧuk−1 + O[k − 2].
2
Consequently, if ϕ is a symmetry, then, since sym(B) is closed under the Jacobi
bracket,
 
1 k−1 dk−1 A
(T1 ) ϕ = −
0 k−1
u1 + O[0]
2 dtk−1
is a symmetry as well. But from (30) one sees that first-order symmetries are
linear in t. Thus, we obtain

Proposition 3 If ϕ = Auk + O[k − 1] is a symmetry of the Burgers equation,

then A is a kth degree polynomial in t.

Final description. Note that direct computations show that the equation B
possesses a third-order symmetry of the form
3 3 3
ϕ03 = u3 + u0u2 + u20 + u20u1 .
2 2 4
Using the actions T22 and T30, one can see that
 k−1
3 k!(k − 1)!
((T22)i ◦ (T30 ◦ T22)k−1 )u1 = − uk + O[k − 1] (31)
2 (k − i)!

is a symmetry, since u1 is the one.

Theorem 4 The symmetry algebra sym(B) for the Burgers equation B = {ut =
uux + uxx}, as a vector space, is generated by elements of the form

ϕik = ti uk + O[k − 1], k ≥ 1, i = 0, . . . , k,

which are polynomial in all variables. For the Jacobi bracket one has
1
{ϕik , ϕjl } = (li − kj)ϕi+j−1
k+l−2 + O[k + l − 3]. (32)
2
The algebra sym(B) is simple and has ϕ01 , ϕ22 and ϕ03 as its generators.

12
Proof. It only remains to prove that all ϕik are polynomials and that sym(B)
is simple. The first fact follows from (31) and from the obvious observation that
coefficients of both T22 and T30 are polynomials.
Let us prove that sym(B) is a simple Lie algebra. To do this, introduce an
order in the set {ϕik } defining
def
Φ k(k+1) +i = ϕik .
2

P
Then any symmetry may be represented as sα=1 λα Φα , λ ∈ R. Let I ⊂ sym(B)
P
be an ideal and Φ = Φs + s−1 i
α=1 λα Φα be its element. Assume Φs = ϕk for some
k ≥ 1 and i ≤ k.
Note now that
X ∂ ∂ ∂
T11 = Dxα (tu1 + 1) − tDx = −t
α≥0 ∂uα ∂u0 ∂x

and
X ∂ ∂
T20 = Dxα (u2 + u0 u1) − Dx2 − u0 Dx − u1 = − .
α≥0 ∂uα ∂t
Therefore,
((T11)k−1 ◦ (T20)i )Φ = cϕ01 ,
where the coefficient c does not vanish. Hence, I contains ϕ01 . But due to (31) the
latter, together with ϕ22 and ϕ03, generates the whole algebra.
Further details on the structure of sym(B) one can find in [3].

3 Recursion operators and nonlocalities.

In fact, we were successful in getting a complete description of symmetry algebra
for the Burgers equation because sym(B) is a very rich algebra. In other cases
this method may not work, e.g. for the KdV equation (??)6 . Below we expose
another method based on computation of recursion operators.
Recursion operators for the Burgers and KdV equations. Informally
speaking, a recursion operator for an evolution equation E

∂u ∂ku
= F (t, x, u, . . . , k ) (33)
∂t ∂x
is an expression of the form
X
s
R= ai ◦ Dxi ◦ bi ,
i=−∞

6
Though it will be helpful anyway to estimate the symmetry algebra.

13
ai , bi being functions of x, t, . . . , uk , . . ., which preserves symmetries of E, i.e.
R : sym(E) → sym(E). Usually, integrable equations possess such operators.
For example (see [5]), the operator
1 1
Dx + u0 + u1 Dx−1
2 2
is a recursion for the Burgers equation, while
2 1
Dx2 + u0 + u1 Dx−1
3 3
is he one for the KdV equation (??).
Having a recursion operator and a starting symmetry ϕ0, one easily obtains
explicit expressions for the whole series ϕk = Rk ϕ0 . Thus, having a method for
computing recursion operators, we get a powerful means to generate symmetries.
Such a method was found in [1, 2] and we discuss it in action below.
Cartan forms. A Cartan form on E ∞ is a form which vanishes on the distri-
bution CE or, equivalently, which anihilates all total derivatives D1 , . . . , Dn on
E ∞ . Cartan forms form a vector subspace C 1 Λ(E ∞ ) in the space Λ1(E ∞ ) of all
differential one-forms on E ∞ . For the case of equation (33), one can choose the
forms
def
ωs = dus − us+1 dx − Dxs (F )dt, (34)
s = 0, 1, . . ., as a basis in C 1 Λ(E ∞ ). It can easily be checked that iDx ωs = iDt ωs =
0.
Below we shall need formulae for Lie actions of total derivatives on the forms
ωs . Let us make necessary computations in the case (33). For Dx one has:

LDx (ωs ) = Dx (dus − us+1 dx − Dx (F )dt) =

d(Dx us ) − (Dx us+1 )dx − us+1 d(Dx x) − Dxs+1 (F )dt − Dxs (F )d(Dx t) =
dus+1 − us+2 dx − us+1 d(1) − Dxs+1 (F )dt − Dxs (F )d(0),

or
LDx (ωs ) = ωs+1 . (35)
To apply Dt to ωs , note first that

[Dx , Dt ] = Dx ◦ Dt − Dt ◦ Dx = 0. (36)

Hence, using (35) and (36), we obtain

LDt (ωs ) = Dt Dxs (ω0 ) = Dxs (Dt ω0 ). (37)

14
But

Dt (ω0 ) = Dt (du0 − u1 dx − F dt) = d(Dt u0 ) − (Dt u1)dx − (Dt F )dt =

∂F ∂F Xk
∂F
dF − (Dx F )dx − (Dt F )dt = dx + dt + dui −
∂x ∂t i=0 ∂ui
! !
∂F Xk
∂F ∂F X k
∂F
+ ui+1 dx − + Dxi )F ) ,
∂x i=0 ∂ui ∂t i=0 ∂ui
or
X
k
∂F
Dt ω0 = ωi .
i=0 ∂ui

Hence ! ! !
X
k
∂F Xk X s
s ∂F
Dt ωs = Dxs ωi = Dxs−α ωi+α . (38)
i=0 ∂xi i=0 α=0 α ∂ui

Action of Cartan forms on symmetries. Let ω ∈ C 1 Λ(E ∞ ) be a Cartan form

and consequently
ω = a0ω0 + a1ω1 + . . . + as ωs , (39)
ai being functions on E ∞ . Consider another function ϕ and the vector field
∞
X ∂
ϕ = Dxi (ϕ) .
i=0 ∂ui

Hence, contracting ϕ with ω, we obtain

iϕ (ω) = a0 iϕ ω0 + a1 iϕ ω1 . . . + as iϕ ωs =

a0ϕ + a1Dx (ϕ) + . . . + as Dxs (ϕ).

Therefore, any ω of the form (39) determines a differential operator

Rω = a0 + a1Dx + . . . + as Dxs . (40)

Theorem 5 [2] Restrict operator (40) onto symmetries of the equation E, i.e. for
def
any ϕ ∈ sym(E) set Rω (ϕ ) = Rω ϕ . Then Rω preserves the algebra sym(E),
i.e. Rω sym(E) ⊂ sym(E), if and only if the form ω satisfies the equation
[1]
`E ω = 0, (41)
[1]
where `E is the extension of the universal linearizaton operator (17) to Cartan
forms.

15
Thus equation (41) may be used to compute recursion operators. We show how
it works using the example of the Burgers equation.
Basic relations for B. In this case we obtain
!
X
s
`[1]
B (ω) = (Dt − u1 − u0 Dx − Dx2 ) ai ωi =
i=0
X
s
(`B (ai )ωi + ai Dt (ωi ) − u0 aiωi+1 − 2Dx (ai)ωi+1 − aiωi+2 ).
i=0

Using (38) one can now reduce (41) to a system of equations on functions
a0, . . . , as . We shall not do it here in the general form, but write down the equa-
tions for s = 2. In this case one obtains

Dx (a1) = 0,
Dx (a0) + Dx2 (a1) + u0 Dx (a1) = u1a1 + Dt (a1),
Dx2 (a0) + u0Dx (a0 ) + u1a0 = Dt (a0 ). (42)

A ”disappointing” result. Now, if one tries to solve (42) or the equation

[1]
`B ω = 0 in its general form (see [1] for the proof), one will get the only solution
ω = a0 ω0 , a0 ∈ R.

Corresponding operator Rω just multiplies symmetries by the constant a0 and

produces no nontrivial recursions.
And this is not unnatural, since the known recursion operator for the Burg-
ers equation contains the term Dx−1 (see above), which could not appear in our
setting.
R
Nonlocal setting. Introduce a new variable u−1 = Dx−1 (u0 ) = u0 dx and set
def

Z
1
Dx u−1 = u0, Dt u−1 = Dt (u0 ) dx = u1 + u20 .
2
Then we get ”new” total derivatives

f =u ∂ f = (u + 1 u2 ) ∂ + D
Dx 0 + Dx , Dt 1 t
∂u−1 2 0 ∂u−1
and the Cartan form
1
ω−1 = du−1 − u0dx − (u1 + u20)dt.
2
Let us solve the equation
è
Bω = 0, (43)
where èB is the operator `B in which ”old” total derivatives Dx , Dt are substituted
f and D
by D f respectively.
x t

16
Proposition 4 [2] Equation (43) possesses two independent solutions
1 1
ω0 , ω = ω1 + u0ω0 + u1ω−1 .
2 2
The operator Rω0 is the identity, while Rω coincides with the classical recursion
operator for the Burgers equation.

Remark 2 Try to compute symmetries in this setting: a very intersting result

awaits you!
General scheme. Below we expose the way of action in the case of a general
evolution equation.
1. Solve the equation
`E ϕ = 0
for low-order ϕ.
2. Solve the equation
`[1]
E ω = 0

for low-order Cartan forms. Usually you will get trivial solutions only.
3. Extend the setting with nonlocal variables. To do this, solve the equation

`∗E ψ = 0, (44)

where `∗E is the formally adjoint7 to `E . The solutions of (44) are generating
functions of conservation laws for the equation E (see [7] for details).
R
4. Enlarge the setting by adding nonlocal variables Dx−1 ψ = ψ dx for any
solution ψ; equation (44) guarantees the new setting to be well-defined.
5. Solve the equation
è ω = 0
E

in the new setting and get the recursion operator.

For integrable equations this scheme works perfectly.
Acknowledgements. I am grateful to the organizers of the 1st Non-Orthodox
School on Nonlinearity & Geometry for the opportunity to give this lecture.

References
[1] I.S. Krasil’shchik, Some new cohomological invariants for nonlinear differ-
ential equations, Diff. Geom. Appl. 2 (1992) 307–350.

[2] I.S. Krasil’shchik and P.H.M. Kersten, Deformations and recursion operators
for evolution equations, In: Pràstaro A., Rassias Th.M. (Eds.), Geometry in
Partial Differential Equations (1993) World Scientific, Singapore.
Ps Ps
7
If ∆ = i=0 ai Dxi , then ∆∗ = i
i=0 (−1) Dx ◦ ai .

17
[3] I.S. Krasil’shchik, Lychagin V.V. and A.M. Vinogradov, Geometry of Jet
Spaces and Nonlinear Differential Equations, Advanced Studies in Contem-
porary Mathematics 1 (1986) Gordon and Breach, New York, London.

[4] I.S. Krasil’shchik and A.M. Vinogradov, Nonlocal trends in the geometry of
differential equations: symmetries, conservation laws, and Bäcklund trans-
formations, Acta Appl. Math. 15 (1989) 161–209.

[5] P.J. Olver, Applications of Lie Groups to Differential Equations, Graduate

Texts in Mathematics 107 (1986) Springer-Verlag, New York.

[6] L.V. Ovsyannikov, Group Analysis of Differential Equations, Academic

Press, New York, 1982.

[7] A.M. Vinogradov, Local symmetries and conservation laws, Acta Appl.
Math. 2 (1984) 21–78.

18