You are on page 1of 42

Home Search Collections Journals About Contact us My IOPscience

Exactly integrable hyperbolic equations of Liouville type

This article has been downloaded from IOPscience. Please scroll down to see the full text article.

2001 Russ. Math. Surv. 56 61

(http://iopscience.iop.org/0036-0279/56/1/R02)

View the table of contents for this issue, or go to the journal homepage for more

Download details:
IP Address: 84.38.9.201
The article was downloaded on 15/06/2010 at 19:01

Please note that terms and conditions apply.


Russian Math. Surveys 56:1 61–101 2001
c RAS(DoM) and LMS
Uspekhi Mat. Nauk 56:1 63–106 DOI 10.1070/RM2001v056n01ABEH000357

Exactly integrable hyperbolic equations of Liouville type

A. V. Zhiber and V. V. Sokolov

Abstract. This is a survey of the authors’ results concerning non-linear hyperbolic


equations of Liouville type. The definition is based on the condition that the chain of
Laplace invariants of the linearized equation be two-way finite. New results include a
procedure for finding the general solution and a solution of the classification problem
for Liouville type equations.

Contents
§1. Introduction 61
§2. Laplace transformations and Laplace invariants 63
§3. Construction of exact solutions 73
§4. Higher symmetries of Liouville type equations 83
§5. Liouville type equations and Miura transformations for evolution
equations 85
§6. The list of Liouville type equations 88
§7. Concluding remarks 93
Bibliography 99

§1. Introduction
One of the most important problems in the theory of partial differential equa-
tions at its early stages was finding and studying equations integrable in closed
form. Eighteenth and nineteenth century mathematicians such as Euler, Lagrange,
Liouville, Laplace, Darboux, Lie, Jacobi, and Goursat developed a variety of tech-
niques for finding exact solutions [1]–[7].
At the beginning of the 20th century, after a revaluation of basic priorities in the
theory of partial differential equations under the very strong influence of mathemat-
ical physics, many classical results on exact integration were essentially forgotten
even by experts in the field. Recently, interest in these results has been increasing
dramatically owing to the discovery of a new fundamental method for exact inte-
gration of non-linear partial differential equations: the inverse scattering method
[8], [9].

This research was supported by the Russian Foundation for Basic Research (grant nos. 99-01-
00431 and 99-01-00294). The second author was also supported by INTAS (grant no. 1782).
AMS 2000 Mathematics Subject Classification. Primary 35L70; Secondary 35A22, 35A30,
35L75, 35Q53, 37K10, 37K35, 37J15, 37K15.
62 A. V. Zhiber and V. V. Sokolov

The Liouville equation


uxy = exp u (1.1)
is a widely known example of an exactly integrable non-linear partial differen-
tial equation. The study of its formal properties has led to various definitions,
generally non-equivalent, of the class of exactly integrable hyperbolic equations
of Liouville type [1], [6], [10]–[16]. In particular, the finiteness of the chain of
Laplace invariants of the linearized equation is chosen as a definition in [17]–[20].
This approach is generalized in [21] to arbitrary second-order equation of the form
F (x, y, u, ux, uy , uxx, uxy , uyy ) = 0.
This definition has proved successful in that it enables one to use several beau-
tiful classical identities [3] related to Laplace invariants. As a result, some general
formulae describing higher symmetries [19] and conservation laws [17], [18] in terms
of Laplace invariants have been obtained (see § 4). Moreover, a unified procedure
for constructing the general solution of a Liouville type equation has been found.
The corresponding results are published for the first time in § 3.
To verify that the chain terminates for a given equation, one need use only
algebraic operations and differentiations. Therefore, the computation can readily
be performed by any computer algebra software.
The termination property of the chain of Laplace invariants can be used as a
basis for constructing a classification of Liouville type equations. Our attempts to
find all Liouville type equations of the form

uxy = F (x, y, u, ux, uy ) (1.2)

have resulted in two new examples of such equations. An explicit formula for the
general solution in these examples is given in § 3.
A general classification result is stated for the first time in § 6. However, the
computations are extremely tedious, and the complete proof of the classification
theorem is about 200 pages long, so there is some small chance of error in the
computations.
We note that the general classification problem for integrable equations (1.2),
which includes describing not only all Liouville type equations, but also equations
integrable by the inverse scattering method, such as the sine-Gordon equation uxy =
sin u, is extremely hard. Although some partial results have been obtained in the
framework of the symmetry approach [22], [23], the study is far from being complete.
There is a recent new example of such an equation [24], [25]. It has the form
 
uxy = S(u) 1 − u2x 1 − u2y , (1.3)

where S is arbitrary solution of the ordinary differential equation

S  − 2S 3 + λS = 0

with an arbitrary constant λ. In general position, S(u) is none other than the
Jacobi elliptic sine. Equation (1.3) is connected with the equation

vxy = c sin(v)
Exactly integrable hyperbolic equations of Liouville type 63

by the differential substitution

v = arcsin(ux ) + arcsin(uy ) + P (u),

where P 2 = 2S  − 2S 2 + λ. The constant c is determined via the value of the first


integral I = S 2 − S 4 + λS 2 by the formula I = c2 .
The degenerate case 
1
uxy = 1 − u2x 1 − u2y (1.4)
u
of (1.3) is an equation of Liouville type. We use it in § 3 as an illustrative example.
In § 7 we present a new integrable equation of the form (1.2) connected with the
Tzitzéica equation by a differential substitution. We also give a list of currently
known equations of sine-Gordon type and conjecture that the list is complete.
The Miura transformation w = ux − 12 u2 , which relates the modified Korteweg–de
Vries equation
ut = uxxx − 32 u2 ux
to the Korteweg–de Vries equation

wt = wxxx + 3wwx,

is well known in the theory of integrable evolution equations.


Such transformations prove to be closely related to hyperbolic equations of Liou-
ville type. In particular, the Miura transformation is generated by the equation
uxy = uuy . The corresponding theory is presented in § 5.
In § 7 the definition of Liouville type equations is generalized to the case of
non-linear hyperbolic systems of the form

uxy = F (x, y, u, ux, uy ). (1.5)

This paper is a survey of the authors’ results obtained in the last six years and
partly published in [19], [20], [25].
The authors are grateful to E. V. Ferapontov, N. Kamran, M. V. Pavlov,
A. B. Shabat, S. Ya. Startsev, and S. P. Tsarev for their interest in this work
and for many useful discussions.

§2. Laplace transformations and Laplace invariants


2.1. Laplace transformations for linear equations.
In classical papers (see [3], [26], [27]), Laplace transformations and Laplace
invariants were defined for a linear hyperbolic equation of the form
 2 
∂ ∂ ∂
+ a(x, y) + b(x, y) + c(x, y) V = 0. (2.1)
∂x∂y ∂x ∂y

One can readily verify that the functions

∂a ∂b
h0 = + ab − c, k0 = + ab − c (2.2)
∂x ∂y
64 A. V. Zhiber and V. V. Sokolov

are invariants under transformations of the form V → α(x, y) V . These functions


are called the principal Laplace invariants of (2.1).
It is easily seen that (2.1) can be rewritten as the system
   
∂ ∂
+ a V = V1 , + b V1 = h0 V. (2.3)
∂y ∂x

If h0 = 0, then the function V1 defined by the first formula in (2.3) satisfies the
equation  2 
∂ ∂ ∂
+ a1 (x, y) + b1 (x, y) + c1 (x, y) V1 = 0, (2.4)
∂x∂y ∂x ∂y
where
a1 = a − (ln h0 )y , b1 = b, c1 = a1 b1 + by − h0 .
Equation (2.4) is called the Y -Laplace transformation of (2.1). The principal
Laplace invariants of (2.4) are given by

h1 = 2h0 − k0 − (ln h0 )xy , k1 = h0 .

If h1 = 0, then there is a Y -Laplace transformation of (2.4), and so on. As a result,


we obtain a chain of equations
 
∂2 ∂ ∂
+ ai (x, y) + bi (x, y) + ci (x, y) Vi = 0, i ∈ N, (2.5)
∂x∂y ∂x ∂y

whose coefficients and Laplace invariants are related by

ai = ai−1 − (ln hi−1 )y , bi = bi−1 , ci = ai bi + by − hi−1 ,


(2.6)
hi = 2hi−1 − hi−2 − (ln hi−1 )xy , ki = hi−1 .

Here a0 = a, b0 = b, and c0 = c.
Let us find the relationship between solutions of (2.1) and (2.5). We rewrite each
equation (2.5) as the system
   
∂ ∂
+ ai Vi = Vi+1 , + b Vi+1 = hi Vi , (2.7)
∂y ∂x

where

ai = a − ln(h0 h1 · · · hi−1 ). (2.8)
∂y
Let  
ψ = exp (− a dy), ψ = exp (− b dx). (2.9)

Then formulae (2.7) can be rewritten as

∂ 1 ∂ 1
V1 = ψ (V ), Vi+1 = ψh0 · hi−1 (Vi ), i ∈ N,
∂y ψ ∂y ψh0 · hi−1
Exactly integrable hyperbolic equations of Liouville type 65

and
ψ ∂ 1
Vi = (Vi+1 ).
hi ∂x ψ
It follows that

∂ 1 ∂ 1 ∂ 1 ∂ 1
Vk+1 = ψh0 · hk−1 ··· (V ), (2.10)
∂y hk−1 ∂y hk−2 ∂y h0 ∂y ψ
ψ ∂ 1 ∂ 1 ∂ 1 ∂ 1
V = ··· (Vk ) (2.11)
h0 ∂x h1 ∂x hk−2 ∂x hk−1 ∂x ψ

for each k ∈ N. Thus if we know how to solve (2.1), then we can solve each of
equations (2.5), and vice versa.
The X-Laplace transformation is defined in a similar way. Namely, (2.1) is
equivalent to the system
   
∂ ∂
+ b V = V−1 , + a V−1 = k0 V. (2.12)
∂x ∂y

def
If the invariant h−1 = k0 for (2.1) is not zero, then V−1 satisfies the equation
 
∂2 ∂ ∂
+ a−1 (x, y) + b−1 (x, y) + c−1 (x, y) V−1 = 0, (2.13)
∂x∂y ∂x ∂y

which is called the X-Laplace transformation of (2.1). By continuing this process,


we obtain a chain of equations
 
∂2 ∂ ∂
+ a−i (x, y) + b−i (x, y) + c−i (x, y) V−i = 0, i ∈ N, (2.14)
∂x∂y ∂x ∂y

whose coefficients and invariants are related by

a−i−1 = a−i, b−i−1 = b−i − (ln h−i−1 )x ,


c−i−1 = a−i−1 b−i−1 + ax − h−i−1 , (2.15)
h−i = 2h−i−1 − h−i−2 − (ln h−i−1 )xy , k−i = h−i−1 .

The counterparts of (2.10) and (2.11) have the form

∂ 1 ∂ 1 ∂ 1 ∂ 1
V−s−1 = ψh−1 · h−s ··· (V ), (2.16)
∂x h−s ∂x h−s+1 ∂x h−1 ∂x ψ
ψ ∂ 1 ∂ 1 ∂ 1 ∂ 1
V = ··· (V−s ). (2.17)
h−1 ∂y h−2 ∂y h−s+1 ∂y h−s ∂y ψ

Definition 1. The set of all principal invariants hi, i ∈ Z, of (2.5) and (2.14) is
called the sequence of Laplace invariants of (2.1).
66 A. V. Zhiber and V. V. Sokolov

According to formulae (2.6) and (2.15), we have ki = hi−1 , i ∈ Z, and hence all
invariants ki also belong to this sequence.
The sequence of Laplace invariants is uniquely determined by the recursion for-
mula
hi = 2hi−1 − hi−2 − (ln hi−1 )xy , i ∈ Z,

and the initial data

∂a ∂b
h0 = + ab − c, h−1 = + ab − c.
∂x ∂y

2.2. Laplace invariants of the linearized equation. For a given solution


u0 (x, y) of (1.2), the formula u = u0 + εV + . . . results in a linear equation for V
with coefficients depending on u0 . Since the linearized equation has the form (2.1),
we can define the corresponding sequence of Laplace invariants.
In what follows, we intend to study properties shared by these invariants for all
solutions of (1.2). To this end, let us consider the more formal version
 
∂F ∂F ∂F
Dx ◦ Dy − Dx − Dy − V =0 (2.18)
∂ux ∂uy ∂u

of the linearized equation (1.2). Here Dx = D and Dy = D stand for the total
differentiations with respect to x and y by virtue of (1.2).
The operators D and D are more complicated than partial differentiations. In
particular, these operators have non-trivial kernels for some equations (1.2); more-
over, the solution of the equation DD(Z) = 0 does not necessarily have the form
Z = W + W , where D(W ) = D(W ) = 0, and so on. Thus, some special care must
be taken in dealing with these operators, and we start from rigorous definitions.
Most notions in the local theory of non-linear differential equations, such as
symmetries, first integrals, conservation laws, or Lax representations, are defined
in terms of some identities that must be valid for each solution u. To verify these
identities, one must eliminate all partial derivatives of u that can be expressed from
the equation and its differential corollaries.
For example, if we are dealing with solutions of (1.2), then uxy is always replaced
by the right-hand side F (x, y, u, ux, uy ), the derivative uxyy is replaced by the
expression
∂F ∂F ∂F ∂F
+ uy + F+ uyy ,
∂y ∂u ∂ux ∂uy
and so on. One can readily see that each mixed derivative of u can thus be expressed
in terms of
x, y, u, u1 = ux, u2 = uxx , u3 = uxxx, . . . ,
(2.19)
u1 = uy , u2 = uyy , u3 = uyyy , . . . .

No relationships between the functions (2.19) can be derived from (1.2) and
its differential corollaries. Therefore, these functions are treated as independent
variables in all definitions and computations.
Exactly integrable hyperbolic equations of Liouville type 67

The operators D and D are derivations of the space F of locally analytic functions
depending on finitely many of the variables (2.19). These derivations are uniquely
determined by the formulae

D(uk ) = uk+1 , D(uk ) = uk+1 , u0 = u0 = u, k = 0, 1, 2, . . . ,


[D, D] = 0, DD(u) = F (x, y, u, u1, u1 ).

They can be rewritten in the form of vector fields as


 ∞  i−1 ∞
∂ ∂ ∂
D= + ui+1 + D (F ) ,
∂x ∂ui ∂ui
i=0 i=1
 ∞  ∞
∂ ∂ ∂
D= + ui+1 + Di−1 (F ) .
∂y i=0 ∂ui i=1 ∂ui

Although at first glance it seems that D is defined in terms of D and vice versa,
the vector fields are actually well defined by these formulae.
Since the definitions of Laplace transformations and Laplace invariants in § 2.1
are purely algebraic and use only the fact that the partial differentiations are com-
muting derivations, we see that these definitions can be generalized in a straight-
forward manner to the case of the linearized equation (2.18). One just replaces the
partial derivatives by the total derivatives D and D by virtue of (1.2).
According to formulae (2.2), the principal invariants (2.18) have the form
 
def ∂F ∂F ∂F ∂F
H1 = h0 = −D + + , (2.20)
∂u1 ∂u1 ∂u1 ∂u
 
def ∂F ∂F ∂F ∂F
H0 = k0 = −D + + . (2.21)
∂u1 ∂u1 ∂u1 ∂u
We note that the functions H1 and H0 are invariant under transformations of the
form u → ψ(x, y, u) in (1.2).
The remaining invariants are defined recursively (compare with (2.6) and (2.15))
by the formulae

DD(ln Hi) = −Hi+1 − Hi−1 + 2Hi, i ∈ Z. (2.22)

One can readily see that all invariants are rational functions of derivatives of
order ≥ 1. For example,

∂2F ∂2F
H1 = αu2 + β, α=− 2 , H0 = αu2 + β, α = − 2 , (2.23)
∂u1 ∂u1
u2 (p1 u3 + p2 u32 + p3 u22 + p4 u2 + p5 ) + p6 u3 + p7 u32 + p8 u22 + p9 u2 + p10
H2 = ,
(αu2 + β)2
(2.24)

where the coefficients α, α, β, β, and pi are certain specific functions of the right-
hand side F of (1.2) and its partial derivatives with respect to the variables x, y,
u, u1 , and u1 .
68 A. V. Zhiber and V. V. Sokolov

In particular,
∂3F ∂3F ∂2F ∂4F
p2 = 3 2 − . (2.25)
∂u1 ∂u1 ∂u1 ∂u21 ∂u31 ∂u1
The structure of H−1 is similar to that of H2 with ui and ui interchanged.
The increasing order of derivatives in the Laplace invariants is due to the fact
that in general the operator ∆ = DD occurring in the chain (2.22) raises the order
of the higher derivatives by one. One can readily see that for i > 0 the maximal
possible order of Hi with respect to u2 , u3 , . . . is equal to i + 1, while the maximal
possible order with respect to u2 , u3 , . . . is equal to i.
Definition 2. Equation (1.2) is called an equation of Liouville type if there are
r ≥ 1 and s ≥ 0 such that Hr = H−s ≡ 0.
In the case of linear equations (2.1), this definition describes the class of equations
with a finite chain of Laplace invariants. Such equations were studied in detail by
classical mathematicians. Formulae for their general solutions, along with other
useful information, can be found in [3]. It was shown in [28] that the chain of Laplace
invariants is finite if and only if the corresponding linear operator is factorable in
a certain extended sense (factorability into operator ideals ‘of order 1’).
Equation (1.1) is the simplest non-linear equation of Liouville type. It follows
from (2.21) and (2.20) that H0 = H1 = exp u for this equation. One can readily
verify that
H2 = H−1 = 0. (2.26)
Equations 4–7 in § 6 are different examples of equations satisfying condition (2.26).
Let us prove the following important assertion (see [29], [23]).
Theorem 1. For each equation (1.2) of Liouville type, there are functions
ψ(x, y, u, u1 , . . . , up) and ψ(x, y, u, u1 , . . . , up ) such that

∂F ∂F
= D ln ψ(x, y, u, u1 , . . . , up ), = D ln ψ(x, y, u, u1 , . . . , up ). (2.27)
∂u1 ∂u1

Lemma 1. Suppose that a function g(x, y, u, u1 , . . . , up, u1 , . . . , uq ), where p > 0,


satisfies the relation

DD(g) = f(x, y, u, u1 , . . . , un , u1 , . . . , um ), n ≤ p. (2.28)

Then  
∂g ∂F ∂g
D =− . (2.29)
∂up ∂u1 ∂up

Proof. Obviously,
 
∂g ∂F ∂g
DD(g) = up+1 D + +··· ,
∂up ∂u1 ∂up

where the dots stand for terms independent of up+1 . Thus if p + 1 > n, then
condition (2.26) holds.
Exactly integrable hyperbolic equations of Liouville type 69

Proof of Theorem 1. If the chain of invariants Hi, i > 0, terminates, then there
is an invariant Hn of maximum order k in the variables ui . Suppose that k > 0.
Then, applying Lemma 1 to the relation
DD(ln Hn ) = 2Hn − Hn+1 − Hn−1
and using (2.26), we obtain the first of the desired formulae with
 −1
∂Hn
ψ= .
∂uk
If all invariants are independent of u1 , u2 , . . . , then it follows from (2.20) that
∂F
∂u1 = 0, and we can take ψ = 1. The second formula in (2.27) can be proved in a
similar way.
In what follows, the functions ψ and ψ play the same role as the functions (2.9)
in formulae (2.10) and (2.11).
2.3. Equations whose Laplace invariants from a chain of length 2. A
Laplace invariant vanishes identically if and only if the right-hand side F of (1.2)
satisfies a certain overdetermined system of partial differential equations. The sim-
plest systems of this sort provide important information about equations of Liouville
type.
Let us consider the class of equations with the termination condition (2.26).
Formula (2.25) shows that if H2 = 0, then
 
∂2 ∂2F
ln = 0. (2.30)
∂u1 ∂u1 ∂u21
Likewise, the condition H−1 = 0 implies that
 
∂2 ∂2F
ln = 0. (2.31)
∂u1 ∂u1 ∂u21
By solving (2.30) and (2.31) simultaneously, we arrive at the following assertion.
Proposition 1. If (1.2) satisfies the conditions (2.26) and
∂ 2F ∂ 2 F
= 0, (2.32)
∂u21 ∂u21
then it has the following structure:
uxy = P (x, y, u, ux)P (x, y, u, uy ) + Q1 ux uy + Q2 ux + Q3 uy + Q4 , (2.33)
where the Qi are some functions of the variables x, y, and u.
Remark 1. Obviously, one can ensure that Q1 = 0 by an appropriate transformation
of the form u → ψ(x, y, u).
Unfortunately, attempts to clarify the structure of P , P , and Qi further by
a straightforward application of condition (2.26) encounter serious computational
difficulties. Nevertheless, a systematic analysis of the system
DD ln H1 = 2H1 − H0 , DD ln H0 = 2H0 − H1 , (2.34)
which is equivalent to (2.26), gives an interesting result. In this subsection, we
obtain a general formula for a differential substitution that relates equations (1.2)
satisfying the assumptions of Proposition 1 to the wave equation vxy = 0.
The right-hand side of the first condition in (2.34) is independent of the third
derivatives and is linear in the second derivatives, while H1 is given by (2.23).
70 A. V. Zhiber and V. V. Sokolov

Proposition 2. Suppose that


DD ln (a1 u2 + a2 ) = b1 u2 + b2 u2 + b3 , (2.35)
where ai and bi are some functions of the variables x, y, u, u1 , and u1 . Then
 
a1 u2 + a2 = q(x, y, u, u1) q(x, y, u, u1 ) u2 + P (x, y, u, u1)
for some q, q, and P . Furthermore,

∂F
= D ln (u2 + P ), (2.36)
∂u1
     
∂2F ∂F ∂ ln q ∂F ∂ ln q
b1 = + D + , b 2 = D + . (2.37)
∂u21 ∂u1 ∂u1 ∂u1 ∂u1

Proof. In general, the left-hand side of (2.35) has the form (2.24). Equating p1 to
zero in this expression, we see that a2 = a1 P (x, y, u, ux) for some P . The equation
p6 = 0 is equivalent to (2.36). (This observation is not obvious at all!) Finally, the
condition p2 = 0 is equivalent to
∂ 2 a1 ∂a1 ∂a1
a1 = ,
∂u1 ∂u1 ∂u1 ∂u1
and we obtain a1 = q(x, y, u, u1 ) q(x, y, u, u1 ). The desired formulae (2.37) for b1
and b2 can be obtained by a straightforward computation.
By applying Proposition 2 to the first condition in (2.34), we obtain
∂2F
− = q(x, y, u, u1)q(x, y, u, u1 ).
∂u21
With regard for the ‘symmetric’ relation
∂2F
− = r(x, y, u, u1)r(x, y, u, u1 ),
∂u21
we arrive at formula (2.33), where

∂2P ∂2P
q= , q = −P , r= , r = −P.
∂u21 ∂u21
It follows from (2.36) that
 2 

∂F ∂ F
= D ln (H1 ) − ln . (2.38)
∂u1 ∂u21
This relation, as well as the symmetric relation
 2 

∂F ∂ F
= D ln (H0 ) − ln (2.39)
∂u1 ∂u21
(compare with (2.27)) is of great importance for the complete classification of (1.2)
and (2.32) with the termination condition (2.26).
Exactly integrable hyperbolic equations of Liouville type 71

Theorem 2. Every equation (1.2) satisfying conditions (2.32) and (2.26) is con-
nected with the wave equation vxy = 0 by the differential substitution
v = Φ(x, y, u, u1, u1 , u2 , u2 ),
where  3
∂2F
H14
∂u21
Φ = ln  3 . (2.40)
∂2F
H04
∂u21
Proof. By subtracting the second equation in (2.34) from the first, we obtain
     
H1 ∂F ∂F
DD ln = 3(H1 − H0 ) = 3D − 3D .
H0 ∂u1 ∂u1
Further, using (2.38) and (2.39), we see that DD(Φ) = 0 for the function Φ defined
in (2.40).
Corollary 1. The functions W = D(Φ) and W = D(Φ) satisfy the conditions
D(W ) = D(W ) = 0.
2.4. Integrals. The existence of functions W and W of the variables (2.19) such
that D(W ) = D(W ) = 0 (see Corollary 1) is a remarkable property of Liouville
type equations.
Definition 3. A function W (x, y, u, u1, . . . , up) is called a Y -integral of the equa-
tion (1.2) if D(W ) = 0. (We assume that W is not a function of x alone.)
The condition D(W ) = 0 implies that by substituting an arbitrary solution
u(x, y) of (1.2) into W we obtain a function depending on x alone. Needless to say,
distinct solutions in general give distinct functions.
Likewise, an X-integral is a function W (x, y, u, u1 , . . . , up ) satisfying the condi-
tion D(W ) = 0.
In some sense, the notion of a Y -integral [16], [19], [20], [22], [23], [25], [30] is
a generalization of the notion of a first integral for ordinary differential equations.
However, we have the following important difference. Every ordinary differential
equation has first integrals, but in general it is not easier to find a non-trivial first
integral than to solve the equation itself. The case of partial differential equations
is quite different: only exceptional equations possess Y -integrals, but if it is already
known that an equation has a Y -integral of a given order, then it is not difficult to
find this integral.
Obviously, Y - and X-integrals can be defined as non-trivial functions of the
variables (2.19) constant along the characteristics. In this sense, they are close to
Riemann invariants, well known in hydrodynamics. However, Riemann invariants
depend only on the unknown functions but not on their derivatives.
It is obvious that if w is a Y -integral and Q is an arbitrary function, then the
expression  
W = Q x, w, D(w), · · · , Dk (w) (2.41)
is also a Y -integral. The converse is also true. More precisely, the following assertion
holds [23], [29].
72 A. V. Zhiber and V. V. Sokolov

Proposition 3. Every Y -integral can be represented in the form (2.41), where w


is some Y -integral of minimum order.
The integral w occurring in Proposition 3 will be called the minimal Y -integral.
It is defined up to an arbitrary transformation of the form w → φ(x, w). Needless
to say, similar statements are also valid for X-integrals.
The minimal integrals of the Liouville equation (1.1) are

w = u2 − 12 u21 , w = u2 − 12 u21 . (2.42)

The existence of integrals can be taken as a definition of Liouville type equations


[2], [16], [30], [31]. The following assertion [17]–[19] shows that this definition is
equivalent to Definition 2.
Theorem 3. Equation (1.2) possesses Y - and X-integrals

W (x, y, u, u1, . . . , up), W (x, y, u, u1 , . . . , up )

if and only if Hr = H−s ≡ 0 for some r and s. If this is the case, then r ≤ p and
s ≤ p − 1.
The following theorem is very important in various issues related to Liouville
type equations (see Definition 2).
Theorem 4. For each equation (1.2) of Liouville type, the coefficients of the linear
differential operator L given by the formula

ψ 1 1 1 1 ψH1 · · · Hr−1
L= H0 H−1 · · · H1−s D D··· D ···D D (2.43)
ψ H1−s H0 H1 Hr−1 ψ
are Y -integrals.
Proof. Since Hr = 0, it follows that formulae (2.7) with i = r − 1 have the form
(we note that hi = Hi+1 by virtue of the index shift)
 
  ∂F
D − D(ln ψH1 · · · Hr−1 ) Vr−1 = Vr , D− Vr = 0. (2.44)
∂u1

We set Vr = 0. Then Vr−1 = ψH1 · · · Hr−1 Q(x) satisfies (2.44) for an arbitrary
function Q(x). It follows from (2.11) that
 
1 1 1 ψ
V =ψ D · · ·D D H1 · · · Hr−1 Q(x) .
H1 H2 Hr−1 ψ

By (2.16), we have

1 1 1 1
V−s−1 = ψH0 H−1 · · · H1−s D D··· D D V. (2.45)
H1−s H−1 H0 ψ

Since  
∂F
D− V−s−1 = 0,
∂u1
Exactly integrable hyperbolic equations of Liouville type 73

it follows that D(ψ−1 V−s−1 ) = 0. By substituting the above expressions for V−s−1
and V into this formula, we obtain DL(Q) = 0. Since Q(x) is arbitrary, it follows
that D(L) = 0. This completes the proof of Theorem 4.
In the same way, one can show that all coefficients of the operator

ψ 1 1 1 1 ψH0 · · · H−s+1
L= H1 · · · Hr−1 D D··· D D··· D (2.46)
ψ H r−1 H 0 H −1 H −s+1 ψ

are X-integrals.
Remark 2. The function ψ is determined up to multiplication by Y -integrals. We
can ensure that the order of ψ is less than the order of the minimal Y -integral;
then ψ is unique up to a factor depending only on x. Owing to the ambiguity
in the choice of ψ, the operator L is defined up to a transformation of the form
L → W LW −1 , where W is an arbitrary Y -integral.
The coefficients of L and L can be found algorithmically with the use of a
computer. As far as the authors know, this is the most efficient method for finding
integrals of a given Liouville type equation.

§3. Construction of exact solutions


3.1. The general scheme. The Liouville equation (1.1) has the well-known
general solution


2Z  (x)Z (y)
u(x, y) = ln , (3.1)
(Z(x) + Z(y))2
where Z(x) and Z(y) are arbitrary functions. By substituting this solution
into (2.42), we obtain
Z  3 Z 2
w=  − . (3.2)
Z 2 Z 2
It is important to us that Z(x) parametrizes the kernel of L. Indeed,

L = exp(u)D exp(−u)D exp(−u)D exp(u) = D3 + (2uxx − u2x )D + uxxx − ux uxx

by (2.43). Thus,
L = D3 + 2wD + w . (3.3)
It is well known that the functions

1 Z Z2
ϕ1 = , ϕ2 = , ϕ3 = , (3.4)
Z Z Z

where Z(x) is connected with w by (3.2), form a basis in the kernel of L.


We claim that the general solution of every Liouville type equation can be
expressed in terms of the kernels of L, L, and their adjoint operators.
Let us consider the function

ψ
A= . (3.5)
ψH1 H2 · · · Hr−1
74 A. V. Zhiber and V. V. Sokolov

∗ ∗
It follows from (2.43) and (2.46) that L(A) = L (A) = 0. Here L is the formal
adjoint of L. Hence

s+r
A= cij ϕi (x)ϕj (y). (3.6)
i,j=1

Here φ1 , φ2 , . . . , φs+r is a basis in the kernel of L, φ1 , φ2 , . . . , φs+r is a basis in the



kernel of L , and the cij are some constants.
Similarly, the function
ψ
A= (3.7)
ψH0 H−1 · · · H1−s
satisfies the equations L(A) = L∗ (A) = 0.
If A is given as a function of x and y, then, using only algebraic operations and
the total derivatives D and D, we can find the values of some other expressions
depending on the variables (2.19). For example, the following lemma shows that
all Laplace invariants of (1.2) can be expressed via the function (3.5).
Lemma 2. Let us define Hi by the recursion relations

Hi = 2Hi+1 − Hi+2 − ln DDHi+1 , i = r − 2, r − 3, . . . , 1 − s,


(3.8)
Hr = 0, Hr−1 = − ln A.

Then
Hi = DDHi, i = r − 1, r − 2, . . . , 1 − s. (3.9)

Proof. By summing the identities

Hi+1 = 2Hi − Hi−1 − DD ln Hi (3.10)

with i = r − 1, r − 2, . . . , 0 and by taking account of the fact that Hr = 0, we obtain

Hr−1 + H1 − H0 − DD ln H1 · · · Hr−1 = 0

or
Hr−1 = DD ln H1 · · · Hr−1 + DFu1 − DFu1 .
Since Fu1 = D ln ψ and Fu1 = D ln ψ by Theorem 1, we have
 
ψ
Hr−1 = DD ln H1 · · · Hr−1 = DD Hr−1 ,
ψ
where
Hr−1 = − ln A.
Let us define Hr−2 by formula (3.8): Hr−2 = 2Hr−1 − ln DDHr−1 . We apply the
operator DD to both sides of this formula. Using (3.10), we obtain

DDHr−2 = 2Hr−1 − DD ln Hr−1 = Hr−2 .

In a similar way, one can prove that DDHr−3 = Hr−3 and so on.
Exactly integrable hyperbolic equations of Liouville type 75

Note that, along with the recursion formula (3.8), we have the following deter-
minant representation of Hi (see [32]):
 
Hr−k = − ln (−1)k(k+1)/2 ∆k+1 (A) , k = 1, . . . , r + s − 1, (3.11)
j−1
where ∆k (A) is the principal k-minor of the matrix Aij = Di−1 D A.
From the practical point of view, we can use these formulae as follows. On the
one hand, we know A, A, D(A), D(A), and the Laplace invariants as functions of
the variables (2.19). On the other hand, formulae (3.6) and (3.8) express them as
functions of x and y. Hence, using the implicit function theorem, we can find u
together with u1 , u1 , . . . as functions of x and y. This provides the desired formula
for the general solution u(x, y). Moreover, one establishes relationships between
the constants cij in (3.6) in the course of the computations. We point out that
these constants are not arbitrary in general.
In each specific case, this scheme can be simplified by various technical tricks.

In particular, if the operators L, L, L∗, and L admit a factorization, then the
corresponding kernels can be described much more easily (see the examples below).
Moreover, it is sometimes more convenient to seek the variables u, u1, u1 , . . . not
from the identities for A, A, D(A), D(A), and Hi, but from some equivalent simpler
identities.
Example 1. This is the simplest example showing how to find the general solu-
tion (3.1) of the Liouville equation (1.1) in a regular way.
It was already mentioned that
∗ 3
L = −L∗ = D3 + 2wD + wx, L = −L = D + 2wD + wy

for the Liouville equation. Let Z(x) be a function generating the kernel of L by

formula (3.4). Likewise, the kernel of L is generated by
2
1 Z Z
ϕ1 =  , ϕ2 =  , ϕ3 =  .
Z Z Z
It follows from (3.5) that A = e−u , and from (3.6) we obtain
2 2 2
c1 + c2 Z + c3 Z + c4 Z 2 + c5 ZZ + c6 Z + c7 Z 2 Z + c8 ZZ + c9 Z 2 Z
e−u =  .
ZZ
To determine the constants ci, we note that H0 = H1 = u and hence condition
(3.11) for k = 1 has the form

A = D(A) D(A) − A DD(A).

Substituting the explicit expression for A into this formula, we obtain the system
of algebraic relations

c1 = c2 c3 − c1 c5 , c2 = 2c3 c4 − 2c1 c7 , c3 = 2c2 c6 − 2c1 c8 ,


c4 = c4 c5 − c2 c7 , c5 = 4c4 c6 − 4c1 c9 , c6 = c5 c6 − c3 c8 ,
c7 = 2c4 c8 − 2c9 c2 , c8 = 2c7 c6 − 2c3 c9 , c9 = c7 c8 − c5 c9
76 A. V. Zhiber and V. V. Sokolov

for the coefficients ci . The solution

c1 = 0, c2 = 0, c3 = 0, c4 = 12 , c5 = 1, c6 = 12 , c7 = 0, c8 = 0, c9 = 0

of this system corresponds to (3.1). One can readily verify that all other solutions
also result in formula (3.1) with Z and Z appropriately redefined.
Example 2. We 
 consider (1.4) as a second example. For brevity, we set β =
1 − ux and β = 1 − u2y . The minimal Y - and X-integrals for (1.4) are given by
2

u2 β u2 β
w= − , w= − ,
β u β u

and the functions ψ and ψ (see (2.27)) can be taken in the form ψ = β and ψ = β.
One can readily verify that H2 = H−1 = 0 for (1.4), as well as for the Liouville
equation. The invariants H0 and H1 have the form

βw βw
H0 = , H−1 = 2 . (3.12)
uβ 2 uβ

A straightforward computation using (2.43) and (2.46) yields

2
L = D(D2 + v1 D + v2 + e2v ), L = D(D + v 1 D + v2 + e2v ),

where v = ln w and v = ln w. The adjoints of L and L are given by


∗ 2
L∗ = −(D2 − v1 D + e2v )D, L = −(D − v1 D + e2v )D.

It follows from (3.5) and (3.7) that

uβ uβ
A= , A= . (3.13)
w w

Thus, the functions (3.13) satisfy



L(A) = L (A) = L(A) = L∗(A) = 0. (3.14)

Moreover, one can readily verify that


2
(D2 + v1 D + v2 + e2v )A = −1, (D − v 1 D + e2v )DA = 0,
2
(3.15)
(D + v1 D + v 2 + e2v )Ā = −1, (D2 − v1 D + e2v )DĀ = 0.

We set
X Y
w = −i , w = −i , (3.16)
X Y
Exactly integrable hyperbolic equations of Liouville type 77

where X(x) and Y (y) are arbitrary functions and i2 = −1. Then X 2 /X  and 1/X 
form a basis in the kernel of the operator D2 + v1 D + v2 + e2v . The kernel of the
operator (D2 − v1 D + e2v )D is generated by 1, P (x), and Q(x), where
1
P  = X, Q = . (3.17)
X
2
Similarly, Y 2 /Y  and 1/Y  form a basis in the kernel of D + v 1 D + v 2 + e2v , while
1, P (y), and Q(y), where
  1
P = Y, Q = , (3.18)
Y
 2 
form a basis in the kernel of D − v 1 D + e2v D.
By applying variation of constants to (3.15), we obtain
   
X2 1 X2 1 1 P 1 X2
A = c1  + c2  P + c3  + c4  Q + − Q,
X X X X 2 X 2 X
    (3.19)
Y2 1 Y2 1 1P 1Y2
Ā = c1  + c2  P + c3  + c4  Q + − Q,
Y Y Y Y 2Y 2Y
where ci and ci are some constants.
By differentiating (3.13), we obtain

wD(A) = uy β − uxβ, wD(Ā) = uxβ − uy β, (3.20)


D(wA) = −uuxw, D(wĀ) = −uuy w. (3.21)
The left-hand sides of these equations are known, as functions of x and y. Let us
express ux, uy , and u in terms of X(x) and Y (y) from these identities.
It follows from (3.20) that wD(A) = −wD(A). By substituting (3.16) and (3.19)
into this formula, we obtain
c1 = −c1 , c2 = −c3 , c3 = −c2 , c4 = −c4 .
Now
Aw Āw D(wA) D(wĀ)
β= , β= , ux = − , uy = −
u u uw uw
2
by (3.13) and (3.21). Since β 2 = 1 − u2x and β = 1 − u2y , we obtain
 2  2
D(Aw) D(Āw)
u2 = A2 w2 − = Ā2 w2 − .
w w
It follows from the second equation in this chain that
c1 = c4 = 0, 4c2 c3 + 1 = 0.
Since X(x) and Y (y) are defined up to proportionality, we can set c2 = 12 and
c3 = − 12 without loss of generality. Now the first equation gives the following
definitive answer:
  
u(x, y) = Q(x) + Q(y) P (x) + P (y) ,
where
 
P  Q = P Q = 1.
78 A. V. Zhiber and V. V. Sokolov

3.2. New examples. Almost all examples of Liouville type equations can be
found in classical papers (see [3]). For example, (1.4) is well known.
The authors’ attempts to find all Liouville type equations have resulted in two
new equations. The first equation has the form
1 1 2
uxy = B 2 (B − 1)B(B − 1)2 + B (B − 1)B(B − 1)2 , (3.22)
6u + y 6u + x
where B = B(ux ) and B = B(uy ) are solutions of the cubic equations
1 3 2
1 3
3B − 12 B 2 = ux, 3B − 12 B = uy . (3.23)
The right-hand side of the second equation, which has the form
1
uxy = b(ux ) b(uy ), (3.24)
u
also contains cube roots of the first derivatives. Namely, the functions b(ux) and
b(uy ) are defined as solutions of the cubic equations

(ux − b)(b + 2ux )2 = 1, (uy − b)(b + 2uy )2 = 1. (3.25)

A formula for the general solution of (3.22) is given in [25]. This formula was
obtained by the method discussed in the present paper. In what follows, we show
how the method works using (3.22) and (3.24) as examples.
The corresponding rather tedious computations are substantially simplified in
view of the fact that in both cases the operator L can be factorized into a product of
two operators whose coefficients are still Y -integrals. Apparently, this factorization
is due to the existence of differential substitutions reducing (3.22) and (3.24) to
simpler Liouville type equations. Explicit formulae for these substitutions are given
in the following.
Example 3. Let us consider (3.24). One can readily verify that H3 = H−2 = 0
for this equation. The function
 
1 p2 3 bp u1 p
w= + +3 + Dp , (3.26)
2 2 2 u u
where
u2 b
p=2 + ,
b u
is a minimal Y -integral of (3.24). For this choice of w (which is defined up to a
change of variables w → φ(x, w)), the coefficients of the operator L given by (2.43)
are especially simple. It turns out that L admits the factorization L = L1 L2 , where
L1 = D3 − 4w D − 2D(w), L2 = D2 + D(ln w) D + D2 (ln w) − w. (3.27)
We note that
1
L∗1 = −L1 , L∗2 = D2 − D(ln w) D − w = w L2 . (3.28)
w
Similar formulae are valid for L.
Exactly integrable hyperbolic equations of Liouville type 79

Since ψ = b and ψ = b for (3.24), we see that the basic relations (3.14) become
   
b ∗ b
L = 0, L = 0,
bH1 H2 bH0 H−1
    (3.29)
b ∗ b
L = 0, L = 0.
bH0 H−1 bH1 H2

The function A defined in (3.5) has the form

u3 p(b + 2u1 )
A= . (3.30)
(u1 − b)w

We set Φ equal to the function Aw, that is,

u3 p(b + 2u1 )
Φ= . (3.31)
(u1 − b)

Furthermore, we introduce the notation

u2
F = (3.32)
(u1 − b)(u1 − b)

and

√ 3 6
G= F D ln F · D ln F − (3.33)
2 F
and define w, p, L, L1 , L2 , A, and Φ by similar formulae.
One can verify that relations (3.29) are equivalent to the system
   
Φ Φ
L1 (F ) = L1 (F ) = 0, L2 = L2 = 3F,
w w (3.34)

L1 (Φ) = L1 (Φ) = G, L∗2 (G) = L2 (G) = 0.

According to the general scheme, we must parametrize the kernels of all the opera-
tors in (3.34) by two arbitrary functions, solve system (3.34), and find closed-form
expressions for the functions A = Φ/w and A = Φ/w.
Before proceeding to this technical computation, let us show that the general
solution of (3.24) can be expressed in terms of F , Φ, and Φ.
Differentiating (3.32), we obtain

2u1 + b 2u1 + b
p = D ln F − , p = D ln F − .
u u

It follows from these relations and formulae (3.25) and (3.31) that

Φ+F
u(b + 2u1 )(u1 − b) = .
DF
80 A. V. Zhiber and V. V. Sokolov

Likewise,
Φ+F
u(b + 2u1 )(u1 − b) = .
DF
Squaring and multiplying these identities, we obtain

(Φ + F )2 (Φ + F )2
u4 (u1 − b)(u1 − b) =
(DF )2 (DF )2

by virtue of (3.25). Now it follows from (3.32) that

F (Φ + F )2 (Φ + F )2
u6 = .
(DF )2 (DF )2

Thus, the general solution of (3.24) can be represented in the form


1/3
(Φ + F )(Φ + F )
u = F 1/6 . (3.35)
DF · DF

Now we return to the system (3.34). We easily verify that the function v =
− ln(F/2) satisfies the Liouville equation vxy = exp(v). In other words, (3.24) is
connected with the Liouville equation by the differential substitution

u2
v = − ln .
2(u1 − b)(u1 − b)

Therefore, it follows from (3.1) that

(Z(x) + Z(y))2
F =  . (3.36)
Z  (x)Z (y)

It is convenient to take Z and Z as the arbitrary parametrizing functions.


Obviously,
3
G = − (X1 Y2 + X2 Y1 ),
2
where
2Z 2 − ZZ  Z 
X1 (x) = , X2 (x) = ,
Z 3/2 Z 3/2
2   (3.37)
2Z − ZZ Z
Y1 (y) =  3/2 , Y2 (y) =  3/2 .
Z Z
The functions X1 and X2 form a basis in the kernel of L∗2 . It follows from (3.28)
that Xw1 and Xw2 form a basis in the kernel of L2 . Finally, a basis for the kernel of
L1 is given by (3.4). Similar formulae hold for the kernels of the operators L1 , L2
and L∗2 .
Exactly integrable hyperbolic equations of Liouville type 81

Now one can readily express Φ and Φ via Z(x) and Z(y). Let P1 , P2 , P3 , Q1 ,
and Q2 be arbitrary solutions of the linear equations

3Z 2 6Z 3
L2 (P1 ) = , L2 (P2 ) = , L2 (P3 ) = ,
Z Z Z (3.38)
3 3
L1 (Q1 ) = − X1 , L1 (Q2 ) = − X2 .
2 2

In a similar way, we define P 1 , P 2 , P 3 , Q1 , and Q2 .


Using the above formulae for F and G, we find from (3.34) that

w 2
Φ=  P1 (x) + P2 (x)Z + P3 (x)Z + X2 Q1 (y) + X1 Q2 (y),
Z (3.39)
w 
Φ =  P 1 (y) + P 2 (y)Z + P 3 (y)Z 2 + Y2 Q1 (x) + Y1 Q2 (x).
Z

The system (3.38) has the simple particular solution

 
wP1 wZP3 Z Z2
Q1 = −ZQ2 , Q2 = − + + + ,
X1 Z  X2 Z  X2 Z 2 X1 Z 2
   
Z2 X1 4 X2 P1 X1 P3
P1 = C1 (x) − , P = − + + ,
X1 Z 
2
w X1 X2 w X1 X2
 
1 X2
P3 = − C2 (x) +  ,
X2 Z w

where
Z 1
C1 (x) = √ , C2 (x) = √ . (3.40)
Z Z
The general solution of system (3.38) contains ten arbitrary constants, which
can be made more precise by the rather laborious substitution of (3.35) into (3.24).
It turns out that without loss of generality all these constants can be assigned zero
values. The final answer is given by the formula


1/3
(C 1 − C1 + C 2 Z + C2 Z) (C 1 − C1 − C 2 Z − C2 Z)
u(x, y) = .
Z+Z

Note that the answer still contains four arbitrary constants occurring implicitly in
(3.40) and in the formulae

 Z  1
C 1 (y) =   , C 2 (y) =   .
Z Z
82 A. V. Zhiber and V. V. Sokolov

Example 4. Here we present some key formulae for (3.22). This equation satisfies
the same termination condition H3 = H−2 = 0 for the Laplace invariants as (3.24).
The minimal Y - and X-integrals w and w will be given in § 6. The functions ψ and
ψ (see (2.27)) are defined by

B 4 (B − 1)2 B 2 (B − 1)4
ψ = u2 − − ,
6u + y 6u + x
4 2
B (B − 1)2 B (B − 1)4
ψ = u2 − − .
6u + x 6u + y

The Laplace invariants of (3.22) have the form


B(B − 1) B −1 B
H1 = − ψ,
B(B − 1) (6u + y)(B − 1)2 (6u + x)B 2

B(B − 1) B−1 B
H0 = − 2 ψ,
B(B − 1) (6u + x)(B − 1)2
(6u + y)B
2(6u + y)(6u + x)
H2 = ψ ψ,
B(B − 1) [(6u + x)B 2 (B − 1) − (6u + y)B(B − 1)2 ]2
2(6u + y)(6u + x)
H−1 = 2 ψ ψ.
B(B − 1) [(6u + y)B (B − 1) − (6u + x)B(B − 1)2 ]2

The operators L and L admit the factorizations

L = D(D + w)(D + w)(D + 2w)(D + 3w),


(3.41)
L = D(D + w)(D + w)(D + 2w)(D + 3w).

The solution A, A of the system (3.14) has the form

1 
A= 3
(6u + x)(B − 1)B 5 (B − 1)3 − (6u + y)BB 3 (B − 1)5 ,

1  5 3

Ā = 3 (6u + y)(B − 1)B (B − 1)3
− (6u + x)BB (B − 1)5
.

It is convenient to parametrize the minimal integrals w and w by arbitrary functions


X(x) and Y (y) as follows:

X  (x) Y  (y)


w= , w= .
X  (x) Y  (y)

We omit technical details and note only that finding the functions A and A from
the system (3.14) in this case is much easier than in Example 3, since all the factors
in (3.41) are of first order and there are no problems in describing the kernels of L,

L, L∗ , and L .
Exactly integrable hyperbolic equations of Liouville type 83

The computations involve functions P (x), Q(x), P (y), and Q(y) connected
with X(x) and Y (y) by the differential identities
 
P  = X 2 , P = Y 2 , Q = X 3 , Q = Y 3 .

In [25], there is a misprint in these formulae.


The general solution of (3.22) has the form

x+y K1
u(x, y) = − + √ ,
12 12 K2

where

K1 = 2(X + Y )3 + 3(x − y)(X + Y )(P − P ) + (x − y)2 (Q + Q),


K2 = (x − y)2 (Q + Q)2 − 2(x − y)(P − P ) [2(P − P )2 − 3(Q + Q)(X + Y )]
+ 4(X + Y )3 (Q + Q) − 3(X + Y )2 (P − P ).

We note that (3.22) is connected with the well-known Liouville type equation

vxy = exp (v) vy .

by the differential substitution


A(A − 1)(1 − A − B)
v = − ln
ψ

§4. Higher symmetries of Liouville type equations


Definition 4. An infinitesimal (local ) symmetry of the equation (1.2) is an arbi-
trary function V (x, y, u, ux, . . . , un , uy , . . . , um ) of the variables (2.19) that satisfies
the linearized equation (2.18).
This definition is equivalent to the condition that the dynamic flow of the evo-
lution equation
uτ = V (x, y, u, ux, . . . , un , uy , . . . , um ) (4.1)
takes all solutions of (1.2) to solutions. Sometimes, the evolution equation (4.1)
itself is called a symmetry.
A symmetry is said to be classical if n, m ≤ 1. In this case, (4.1) can be
integrated by the method of characteristics, and the solution is a one-parameter
group of point or contact transformations acting on the set of all solutions of (1.2).
The variable τ plays the role of the group parameter.
Otherwise, the symmetry is called a higher symmetry.
There are two well-known classes of non-linear integrable equations (1.2). The
first class consists of equations whose properties are similar to those of the Liouville
equation (1.1). These equations are integrable in quadratures. The sine-Gordon
equation uxy = sin u is the best-known equation of the second class, which consists
of equations integrable by the inverse scattering method.
84 A. V. Zhiber and V. V. Sokolov

Equations of both types possess higher symmetries. Therefore, the problem of


describing all integrable cases can be stated rigorously as the classification problem
for equations of the form (1.2) possessing higher symmetries [33]–[36]. The latter
problem is very complicated; so far, it has been solved only in several special cases
(see [29]). For example [22], the non-linear Klein–Gordon equation

uxy = F (u)

has higher symmetries if and only if it is equivalent to the Liouville equation, the
sine-Gordon equation, or the Tzitzéica equation

uxy = exp (u) + exp (−2u).

It was shown in [22] that all higher symmetries of the Liouville equation are
described by the formula

V = (D + ux ) G(w, wx, . . . , wk1 ) + (D + uy ) G(w, wy , . . . , wk2 ), (4.2)

where w and w are the minimal Y - and X-integrals given by (2.42).


We note two specific features of formula (4.2). First, each symmetry is a sum
of two symmetries depending only on u, ux, uxx, . . . and only on u, uy , uyy , . . . ,
respectively. For the general equation (1.2) a weaker but similar statement was
proved in [29].
Second, the right-hand side of (4.2) contains arbitrary functions G and G. The
evolution equation uτ = V is not exactly integrable for general G and G. Note
that the supply of higher symmetries for hyperbolic equations of sine-Gordon type
is much more modest: the vector space of all symmetries of order bounded by
an arbitrary constant is finite-dimensional. However, each higher symmetry is an
evolution equation integrable by the inverse scattering method.
Formula (4.2) can be generalized to arbitrary Liouville type equations.
Theorem 5. Suppose that Hr = H−s = 0 for an equation (1.2) of Liouville type.
Let M and M be the differential operators

1 1 1 ψH1 · · · Hr−1
M=ψ D ···D D ,
H1 H2 Hr−1 ψ
(4.3)
1 1 1 ψH0 · · · H1−s
M=ψ D ···D D .
H0 H−1 H1−s ψ

Then the expression


V = M (W ) + M (W ) (4.4)
is a symmetry of (1.2) for any Y - and X-integrals W and W .
Proof. The functions Vr = 0 and Vr−1 = ψH1 · · · Hr−1 W satisfy the system (2.44).
Hence the function V given by (2.11) satisfies (2.18). We readily see that V =
M(W ). Therefore V = M(W ) is a symmetry. For the same reason, V = M(W ) is a
symmetry. To complete the proof, it remains to note that the set of all symmetries
is a vector space.
Exactly integrable hyperbolic equations of Liouville type 85

Remark 3. If H0 = 0, then formula (4.3) does not apply to the operator M. In this
case M coincides with the operator of multiplication by ψ. By the same pattern, if
H1 = 0, then M = ψ.
Using (2.12), one can readily prove that the conditions

V−s−1 = ψW1 , Vr = ψ W 1

hold for an arbitrary symmetry V , where W1 and W 1 are some Y - and X-integrals,
respectively. It follows from (2.43) and (2.46) that the symmetry given by (4.4)
satisfies
V−s−1 = ψL(W ), Vr = ψL(W ).
We see that if V has the form (4.4), then W1 ∈ Im L and W 1 ∈ Im L. The last
conditions have the following meaning. By Proposition 3, every Y -integral is a
function of x, the minimal Y -integral w, and its total derivatives w1 = D(w), w2 =
D2 (w), . . . . Since all coefficients of the operator L are Y -integrals by Theorem 4, it
follows that this operator acts on the set of all functions of x, w, w1, w2 , . . . . Such
a function need not belong to Im L in general. One can show that a necessary and
sufficient condition for a symmetry to be given by formula (4.4) is that W1 ∈ Im L.

§5. Liouville type equations and Miura


transformations for evolution equations
Differential substitutions of Miura type are well known in the theory of integrable
evolution partial differential equations (for example, see [37]–[39]). A relation

v = P (x, u, ux, . . . ) (5.1)

is called a differential substitution from the equation

uτ = f(x, u, ux, . . . )

to the equation
vτ = g(x, v, vx , . . . )
if for each solution u(x, t) of the first equation the function (5.1) satisfies the second
equation. It is certainly not for every function P (x, u, ux, . . . ) that there is a pair of
evolution equations connected by the differential substitution (5.1). On the other
hand, in all known examples, if there is at least one pair with this property, then
there are infinitely many pairs of equations connected by the same substitution.
The problem of a complete classification of all differential substitutions is very
complicated and has not yet been solved in the general setting. However, there has
been substantial progress due to a close relationship between differential substitu-
tions and hyperbolic equations of Liouville type [39].
We claim that the minimal Y - and X-integrals of an arbitrary Liouville type
equation specify differential substitutions of the form (5.1). It is quite likely that
all differential substitutions can be obtained by this construction from hyperbolic
equations of Liouville type, but so far we have no idea how to prove that.
86 A. V. Zhiber and V. V. Sokolov

First, let us consider the Liouville type equation

uxy = uuy . (5.2)

One can readily verify that

ψ = 1, ψ = uy , H0 = 0, H 1 = H 2 = uy , H3 = 0

for this equation. The minimal Y - and X-integrals are given by the formulae

1 uyyy 3 u2yy
w = ux − u2 , w= − . (5.3)
2 uy 2 u2y

It follows from (4.3) and (2.43) with the use of Remark 3 that

M = D2 + uD + ux , M = uy , L = D3 + 2wD + wx . (5.4)

The symmetries
3
uτ = Mw = uxxx − u2 ux (5.5)
2
and
3 u2yy
uτ = Mw = uyyy − (5.6)
2 uy
of (5.2) are well-known exactly integrable evolution partial differential equations.
One can readily verify that the derivative wτ by virtue of (5.5) coincides with
wxxx + 3wwx. In other words, the Y -integral (5.3) of (5.2) specifies a differential
substitution from the mKdV equation (5.5) to the Korteweg–de Vries equation.
This substitution is none other that the famous Miura transformation.
In a similar way, the X-integral specifies a differential substitution from (5.6) to
the KdV equation wτ = wyyy + 3wwy .
Moreover, for each symmetry
 
uτ = D2 + uD + ux G(x, w, wx, . . . , wn ) (5.7)

of (5.2) the Y -integral w = ux − 12 u2 satisfies some evolution equation of the form


wτ = Q(x, w, wx, . . . ). This is due to the fact that the derivation Dτ by virtue of
(5.7) commutes with D and hence takes Y -integrals to Y -integrals. In particular,
Dτ (w) is a Y -integral of (5.7). By Proposition 1, each Y -integral can be represented
in the form (2.41). Hence Dτ (w) = Q(x, w, wx, . . . ).
Let us write out an explicit expression for Q:

wτ = (D − u) uτ = (D − u) (D2 + uD + ux ) G(x, w, wx, . . . , wn )


= (D3 + 2wD + wx ) G(x, w, wx, . . . , wn) = L G(x, w, wx, . . . , wn ).

Similar considerations prove the following assertion.


Exactly integrable hyperbolic equations of Liouville type 87

Proposition 4. Let w be the minimal Y -integral of a Liouville type equation, and


let
uτ = MG(x, w, wx, . . . ) (5.8)
be a symmetry of this equation. Then the symmetry (5.8) is connected with some
evolution equation
vτ = Q(x, v, vx, . . . ) (5.9)
by the differential substitution
v = w(x, u, ux, . . . , uk ). (5.10)
The ambiguity w → f(x, w) in the choice of the minimal integral corresponds to a
point transformation of (5.9).
We readily see that the right-hand side Q of (5.9) has the form
 
Q = w∗ M G(x, v, vx, . . . ) . (5.11)
Here

k
∂w i
w∗ =
def
D
∂ui
i=0

is the Fréchet derivative of w. Since Dτ (w) is a Y -integral for an arbitrary


function G, it follows that all coefficients of the differential operator w∗M are
Y -integrals.
Remark 4. In all known examples, w∗ M coincides (under consistent definitions of
w and ψ) with the operator L given by (2.43). If this is the case, then it follows
from (2.43) and (4.3) that the formula
ψ 1 1 1
w∗ = H0 H−1 · · · H1−s D D··· D (5.12)
ψ H1−s H0 ψ
must be valid. The authors can rigorously prove this formula under the assumption
that the order k of the minimal integral is equal to ord L − ord M.
Let us consider an arbitrary first-order differential substitution
v = P (x, u, ux). (5.13)
To (5.13) we assign the hyperbolic equation
Pu
uxy = − uy , (5.14)
Pux
for which P is a Y -integral. For the Miura transformation, (5.14) coincides
with (5.2). It follows from (2.21) that H0 = 0 for every P . Surprisingly, for
all known substitutions of the form (5.13) the corresponding equations (5.14) are
equations of Liouville type (that is, Hr = 0 for some r > 0).
For example, the well-known differential substitution v = ux +exp (u)+exp (−u)
gives rise to the equation
 
uxy = exp (−u) − exp (u) uy , (5.15)
for which H3 = 0.
88 A. V. Zhiber and V. V. Sokolov

Theorem 6. Equation (5.14) is a Liouville type equation if and only if there is a


change of variables u → f(x, u) such that the function P (x, u, ux) for the equation
obtained by this transformation satisfies

ux = α(x, P )u2 + β(x, P )u + γ(x, P ) (5.16)

with some α, β, and γ. The resulting equation has the X-integral (not necessarily
minimal )
uyyy 3 u2yy
W = − .
uy 2 u2y

Thus, there is a family of substitutions (5.13) depending on three arbitrary


functions α, β, and γ. Apparently, this family describes (up to changes of variables
u → f(x, u)) all possible first-order differential substitutions. In particular, the
function P for (5.15) satisfies the equation ux = −u2 + P u − 1 after the change of
variables u → ln u; for the Cole–Hopf transformation v = ux /u, which linearizes
the Burgers equation vt = vxx + 2vvx , we have ux = P u, and so on.

§6. The list of Liouville type equations


6.1. The list. In this subsection we give a list of non-linear Liouville type equa-
tions and their Y - and X-integrals. The list is complete up to the involution x ↔ y
and (generally complex) transformations of the form

x → ζ(x), y → ξ(y), u → θ(x, y, u). (6.1)

Class 1. Equations of this class have the form

Wy
uxy = − . (6.2)
Wux

Here the function W (x, y, ux) is determined from the equation


n
ux = q0 (x, y) + αi (y) qi (x, W ),
i=1

where αi and qi are arbitrary functions. The function w = W (x, y, ux ) is a Y -


integral. Let
∂n 
n−1
∂i
L= n + ci (y) i
∂y ∂y
i=0

be the uniquely determined differential operator whose kernel is generated by α1 (y),


α2 (y), . . . , αn (y). Then the function


n−1
w = un + ci (y)ui + α(x, y),
i=0
Exactly integrable hyperbolic equations of Liouville type 89

where α(x, y) is determined from the relation


∂α
= −L(q0 ),
∂x
is an X-integral (not necessary minimal).
Class 2 (see § 5).
Pu
uxy = − uy . (6.3)
Pux
Here
ux = α(x, P )u2 + β(x, P )u + γ(x, P ),
where α, β, and γ are arbitrary functions. The integrals are w = P (x, u, ux) and

u3 3 u22
w= − .
u1 2 u21
The following is a well-known example of an equation of this class.
Equation 1.
uxy = eu uy .

Class 3.

uxy = An (x, y) ux uy .
Here An (x, y) is an arbitrary solution of the equation

hn = 0, n ≥ 1,

where the left-hand side hn is determined by the recursion formulae

∂2
hk+1 = 2hk − hk−1 − ln hk , k = 0, 1, . . . , n − 2, (6.4)
∂x∂y
1 1 ∂2
h0 = A2n , h1 = A2n − ln An . (6.5)
4 4 ∂x∂y
In particular, Class 3 contains all equations with
2nλ
An =
λ(x + y) − xy
as well as their degenerate forms with
2n
An = .
(x + y)

The Y -integral w = w(x, y, u, u1, . . . , un) has the form


1     √
w= D − D(h1 ) · · · D − D(h1 · · · hn−1 ) An ux .
An

The integral w can be obtained from w by the replacement D → D, ux → uy .


90 A. V. Zhiber and V. V. Sokolov

Equation 2.
uxy = eu .
The minimal integrals are given by

w = u2 − 12 u21 , w = u2 − 12 u21 .

Equation 3.  
1 1
uxy = + ux uy .
u−x u−y
The minimal integrals are

u2 1 − 2u1 u2 1 − 2u1
w= + , w= + .
u1 u−x u1 u−y

Equation 4.
uxy = f(u)b(ux ),
where
ff  − f 2 = 0, bb + u1 = 0.
The minimal integrals are given by

u2 b f 2
w= + f , w = u2 − u + f 2 du.
b f 2f 1

Equation 5.
uxy = f(u) b(ux ) b(uy ),
where

(ln f) − f 2 = 0, bb + u1 = 0, b b + u1 = 0.
The minimal integrals have the form

u2 b u2 b
w= + f , w= + f .
b f b f

Equation 6.
1
uxy = b(ux ) b(uy ),
u
where

bb + cb + u1 = 0, b b + cb + u1 = 0,
and c is an arbitrary constant. The minimal integrals are

u2 b u2 b
w= − , w= − .
b u b u
Equation 7.
1
uxy = ,
(x + y) b(u1 ) b(u1 )
Exactly integrable hyperbolic equations of Liouville type 91

where
 3 2
b = b3 + b2 , b = b +b .
The minimal integrals have the form

1 1
w = b u2 − , w = b u2 + .
b (x + y) b (x + y)

Equation 8.
1
uxy = b(ux ) b(uy ),
u
where

bb + b − 2u1 = 0, b b + b − 2u1 = 0.
The minimal integrals are

u3 2(b − u1 ) 2 2(2u1 + b) b(u1 + b)


w= + 3
u2 + u2 + ,
b b ub u2
u3 2(b − u1 ) 2 2(2u1 + b) b(u1 + b)
w= + 3 u2 + u2 + .
b b ub u2

Equation 9.
 

1 1 1 1 1
uxy = BB + − B− B ,
Bu1 B u1 6u + x 6u + y 6u + y 6u + x

where
1 3 1 2 1 3 1 2
B − B = u1 , B − B = u1 .
3 2 3 2
The minimal integrals are given by


B 4 (B − 1)2 B 2 (B − 1)4
w = D ln u2 − − − ln B(B − 1)
6u + y 6u + x
 

1 1 1
− + B− B(B − 1),
6u + y 6u + x 6u + x
 4 2

B (B − 1)2 B (B − 1)4
w = D ln u2 − − − ln B(B − 1)
6u + x 6u + y
 

1 1 1
− + B− B(B − 1).
6u + y 6u + x 6u + x

6.2. Comments. 1. For equations of Class 3, it follows from (6.5) that DD ln h1 =


2h1 −2h0 . Hence, the relations (6.4) and (6.5), together with hn = 0, form the open
Cn -Toda lattice, and therefore all possible functions An can be found explicitly [40],
[41].
2. The list shows that the integers r and s in the termination condition Hr =
H−s = 0 can be arbitrarily large for equations of Class 3. The classification of
92 A. V. Zhiber and V. V. Sokolov

equations with large r and s involves substantial difficulties. The main observation
enabling one to overcome these difficulties was published in [25].
3. Needless to say, all ordinary differential equations for the functions f(u), b(ux ),
and b(uy ) (see Equations 5–8) can readily be solved in closed form. However, when
doing that we must consider some particular solutions separately, and so the list
is actually larger. For example, Equation 4 is in fact equivalent to one of the two
equations 
uxy = exp (u) 1 − u2x
and 
uxy = 1 − u2x.
4. To reduce Liouville type equations to one of the canonical forms, we have used
complex transformations. The list of real canonical forms is somewhat broader.

5. Goursat [42] observed that the substitution v = uy reduces the equation

uxy = A(x, y) ux uy (6.6)
to the linear equation
Ay A2
vxy = vx + v. (6.7)
A 4
One can show that (6.6) is of Liouville type if and only if (6.7) is.
The functions hi in the description of Class 3 equations are none other than the
Laplace invariants of (6.7). Every Liouville type equation (6.7) has Y -integrals of
the form

m
w= Mi (x, y) v i .
i=0

The corresponding integrals of (6.6) can be obtained by the substitution v = uy .
6. Roughly speaking, Liouville type equations are equations connected with the
equation vxy = 0 by various substitutions. However, there are non-linear equations
(1.2) connected by substitutions, say, with the equation
vxy = v. (6.8)
In particular, the equation

uxy = 2 ux uy (6.9)
is connected with (6.8) by the substitution
√ √
v = c1 ux + c2 uy ,
where the ci are arbitrary constants.
Equation (6.8), as well as (6.9), is not a Liouville type equation. Thus the notion
of linearizable equations is broader than the notion of Liouville type equations.
7. There are equations for which the chain of Laplace invariants is only one-way
finite. For example, for the equation
uxy = exp (u) b(ux), (ux − b)(b + 2ux )2 = 1,
one has H−1 = 0, but Hi = 0 for i > 0. We do not know if this equation is
integrable in any sense.
Exactly integrable hyperbolic equations of Liouville type 93

§7. Concluding remarks


7.1. One property of the operator L. Let us consider the set G of all functions
depending on the variables

x, u0 = u, u1 = ux , u2 = uxx , ... .

For each f(x, u, u1 , . . . , un ), we set


n
∂f i
f∗ =
def
D.
∂ui
i=0

Then the bracket


[f, g] = g∗ (f) − f ∗ (g)
def
(7.1)
defines the structure of a Lie algebra on G. The bracket (7.1) corresponds to the
commutator of the flows determined by the evolution equations ut1 = f and ut2 = g.
Therefore, the set of all symmetries

uτ = S(x, u, u1 , u2 , . . . ) (7.2)

of a given Liouville type equation is a Lie subalgebra of G. (As usual, we identify


symmetries with their right-hand sides.)
The operator L = D3 + 2uD + u1 corresponding to the Liouville equation
(see Example 1) has the following remarkable property. Its range is a Lie sub-
algebra of G. Namely, one can readily verify that
 
[L(f), L(g)] = L D(f)g − D(g)f + g∗ L(f) − f ∗ L(g) .

Since L is injective, it follows that the last formula specifies the new Lie bracket

[f, g]1 = D(f)g − D(g)f + g∗ L(f) − f ∗ L(g)


def
(7.3)

on G.
For all Liouville type hyperbolic equations known to the authors, the operator
L given by (2.43) has the same property. We shall show that this is quite natural
in view of the results of § 5.
Let Gw be the Lie algebra of functions of the variables x, w, w1, . . . with respect to
the bracket (7.1). It follows from Theorem 4 and Proposition 3 that all coefficients
of L are functions of x, w, w1, . . . , where w is the minimal Y -integral. Thus we have
a differential operator L : Gw → Gw . We claim that under certain assumptions Im L
is a Lie subalgebra of Gw .
Indeed, let us consider the set S of all symmetries (5.8) of a given Liouville type
equation. In all known examples, S coincides with the set of all symmetries (7.2)
and hence is a Lie subalgebra of G with respect to the bracket (7.1). As shown in
§ 5, the symmetry (5.8) is taken by the substitution (5.10) to the equation wτ = Q,
where  
Q = w∗ M G(x, w, w1, . . . ) . (7.4)
94 A. V. Zhiber and V. V. Sokolov

Hence the set of all functions of the form (7.4) is a subalgebra with respect to the
bracket (7.1). To complete the argument, we recall that w∗ M = L for all known
Liouville type equations by Remark 4.
It is not very difficult to verify whether the range of a given differential operator
L with coefficients in Gw is a Lie subalgebra. This condition proves to be rather
stringent.
For example, let us consider operators ‘similar’ to (3.41). Namely, the coefficients
of (3.41) are polynomials in w, w1, w2 , . . . . The operator (3.41) is homogeneous if
we assign the weight 1 to D and the weight i + 1 to wi. Below we give a complete
list of homogeneous operators L of orders 2–6 such that Im L is a Lie subalgebra
of Gw . All these operators admit decompositions into first-order factors.
Second-order operators:
(2)
L1 = D(D + w).
Third-order operators:
(3)
L1 = D(D + w)(D + w).
Fourth-order operators:
(4)
L1 = D (D + w) (D + w) (D + w),
(4)
L2 = D (D + w) (D + w) (D + 2w).

Fifth-order operators:
(5)
L1 = D (D + w) (D + w) (D + w) (D + w),
(5)
L2 = D (D + w) (D + w) (D + 2w) (D + 3w).

Sixth-order operators:
(6)
L1 = D (D + w) (D + w) (D + w) (D + w) (D + w),
(6)
L2 = D (D + w) (D + w) (D + w) (D + w) (D + 2w),
(6)
L3 = D (D + w) (D + w) (D + 2w) (D + 3w) (D + 3w),
(6)
L4 = D (D + w) (D + w) (D + 2w) (D + 3w) (D + 4w).
(5)
We note that the operator L2 coincides with (3.41). It would be of interest to find
all such operators of order > 6 and reveal the beautiful mathematics hidden in the
(j)
sequences of positive integer coefficients in the factors of the operators Li .
One can show that the range of every Hamiltonian differential operator H : G →
G (see [43]) is a subalgebra in G. Therefore, the condition discussed above gives rise
to a non-skew-symmetric generalization of the notion of Hamiltonian operators.
7.2. Liouville type systems. Open Toda lattices associated with Cartan matri-
ces of simple Lie algebras [41], [30] are a straightforward generalization of the Liou-
ville equation to the case of systems of the form (1.5). One of the several equivalent
forms of these lattices is  j
(ui )xy = Ai exp (uj ),
j
Exactly integrable hyperbolic equations of Liouville type 95

where Aji are the entries of the Cartan matrix of the corresponding simple Lie
algebra. It is well known that these systems have Y - and X- integrals [44].
Obviously, most of the definitions, constructions, and statements given in this
paper for Liouville type equations can be generalized to the case of systems (1.5).
However, one encounters a serious problem related to the definition of Laplace
invariants.
The linearization operator (2.18) becomes an operator of the form DD + aD +
bD +c with matrix coefficients. The straightforward generalization of all definitions
to the matrix case is as follows [45]. The principal Laplace invariants are defined
by the formulae

H1 = D(a) + ba − c, H0 = D(b) + ab − c,

and the matrices Hi for i > 1 are determined recursively from the system

DHi − Hi ai−1 + ai Hi = 0, (7.5)


Hi+1 = 2Hi + D(ai − ai−1 ) + [b, ai − ai−1 ] − Hi−1 , (7.6)

where a0 = a. Obviously, in the scalar case these formulae coincide with the
corresponding equations in (2.6).
Suppose that the matrices Hi, i ≤ k, and ai , i ≤ k − 1, are already known.
In this case we determine ak from (7.5) and then find Hk+1 from (7.6). However,
if det Hk = 0, then ak either does not exist at all or is defined not uniquely but
only up to a matrix α such that αHk = 0. In the latter case, the existence and
properties of subsequent Laplace invariants depend essentially on the choice of α.
The degeneration det Hk = 0 for some k is typical of open Toda lattices.
Example 5. Consider the A2 -Toda lattice

uxy = −2 exp u + exp v, vxy = exp u − 2 exp v.

The linearization operator has the form


 
2 exp u − exp v
DD+ .
− exp u 2 exp v

Using (7.5) and (7.6), we obtain


 
1 −4uy + vy −2uy + 2vy
a1 = ,
3 2uy − 2vy uy − 4vy
 
exp u − 2 exp v 2 exp u − exp v
H2 = .
− exp u + 2 exp v −2 exp u + exp v

We see that det H2 = 0.


One can readily prove the following general statement [45] (compare with
Theorem 3).
96 A. V. Zhiber and V. V. Sokolov

Theorem 7. If the system (1.5) has Y - and X-integrals


W (x, y, u, u1, . . . , up ), W (x, y, u, u1 , . . . , up ),
then det Hr = det H−s = 0 for some r ≤ p and s ≤ p − 1.
Thus, for systems of Toda type some of the Laplace invariants are necessarily
degenerate, and we encounter the problem of finding a well-defined sequence of
Laplace invariants.
We set
Zk = Hk Hk−1 · · · H1 . (7.7)
It follows from (7.5) that
Zk (D + a) = (D + ak ) Zk . (7.8)
It is the sequence Zk (rather than Hk ) that is well defined for known Liouville
type systems. Furthermore, if the matrices Hi, i ≤ k (and hence Zk ) and ai ,
i ≤ k − 1, are already known, then we find ak from (7.8) and then determine Hk+1
from the equation
Hk+1 = (D + b)ak − (D + ak )b + Hk , (7.9)
which is equivalent to (7.6). The matrix ak is determined up to an arbitrary matrix
α such that αZk = 0.
Theorem 8. Suppose that the Hi, i ≤ k, are already known and the conditions
(D + a)(ker Zi ) ⊂ ker Zi (7.10)
and
(D − bT )(ker ZiT ) ⊂ ker ZiT (7.11)
are satisfied for i < k. Then ak exists if and only if condition (7.10) is satisfied for
i = k. Furthermore, Zk+1 is independent of the arbitrary matrix α occurring in the
determination of ak if and only if condition (7.11) holds for i = k.
The idea of the proof. It follows from (7.8) that condition (7.10) with i = k is
necessary for the existence of ak . The sufficiency follows from the Kronecker–
Capelli theorem, and (7.9) implies (7.11).
Remark 5. If a = b = 0, then conditions (7.10) and (7.11) are satisfied provided
that the vector spaces ker Zk and ker ZkT admit bases consisting of constant vectors.
Example 6. For the open A3 -Toda lattice
(u1 )xy = 2 exp u1 − exp u2 ,
(u2 )xy = − exp u1 + 2 exp u2 − exp u3 ,
(u3 )xy = − exp u2 + 2 exp u3 ,
all the matrices Zk are uniquely determined, and rank Zk = 4 − k. In particular,
Z4 = 0. The vector e1 = (1, 1, 1)T forms a basis of ker Z2 . The vectors e1 and
e2 = (1, 0, −1)T from a basis in ker Z3 . Bases in ker Z2T and ker Z3T are formed by
the vector f1 = (3, 4, 3)T and the vectors f1 and f2 = (1, 0, −1)T , respectively. We
see that the vector spaces ker Zk and ker ZkT for this Toda lattice admit constant
bases, and therefore, conditions (7.10) and (7.11) hold.
Exactly integrable hyperbolic equations of Liouville type 97

Example 7. For the C3 -Toda lattice

(u1 )xy = 2 exp u1 − exp u2 ,


(u2 )xy = − exp u1 + 2 exp u2 − exp u3 ,
(u3 )xy = −2 exp u2 + 2 exp u3 ,

the matrices Zk are uniquely determined and satisfy rank Z1 = 3, rank Z2 = 2,


rank Z3 = 2, rank Z4 = 1, rank Z5 = 1, and Z6 = 0. All the vector spaces ker Zk
and ker ZkT admit constant bases.
Example 8. For the D3 -Toda lattice

(u1 )xy = 2 exp u1 − exp u2 − exp u3 ,


(u2 )xy = − exp u1 + 2 exp u2 ,
(u3 )xy = − exp u1 + 2 exp u3 ,

one has rank Zk = 4 − k. All vector spaces ker Zk and ker ZkT admit constant bases.
Remark 6. We see from these examples that the indices i for which Zi exhibits a
rank drop coincide with the exponents of the corresponding simple Lie algebra, and
the index h such that Zh = 0 is equal to the Coxeter number.
The termination condition for the sequence Zi can be used as a basis of the
definition of Liouville type systems (1.5).
Definition 5. A system (1.5) of hyperbolic equations is called a system of Liouville
type if conditions (7.10) and (7.11) are satisfied and there are indices r ≥ 1 and
s ≥ 0 such that Zr = Z−s ≡ 0.
Using this definition, let us classify all Liouville type chains of the form
(u1 )xy = 2 exp u1 + k1 exp u2 ,
(7.12)
(u2 )xy = k2 exp u1 + 2 exp u2

with non-singular (k1 k2 = −4) non-diagonal Cartan matrix.


One can readily verify that rank Z1 = 2 and rank Z2 = 1 for any k1 and k2 .
Furthermore, Z3 = 0 if and only if k1 = k2 = −1 (the A2 -Toda lattice).
Let Z3 = 0; then condition (7.10) with i = 3 is satisfied if and only if k1 = −1
or k2 = −1. Without loss of generality, we set k1 = −1. Then Z4 = 0 if and only
if k2 = −2 (the C2 -Toda lattice).
If k2 = −2, then rank Z5 = 1 and condition (7.10) with i = 5 holds if and only
if k2 = −3. In this case, Z6 = 0 (the G2 -Toda lattice).
Thus, we have proved that all Liouville type systems (7.12) are exhausted by the
Toda lattices corresponding to simple Lie algebras of rank 2.
7.3. Equations of sine-Gordon type. Along with the sine-Gordon equation

uxy = sin u,

the Tzitzéica equation


uxy = exp (u) + exp (−2u),
98 A. V. Zhiber and V. V. Sokolov

and (1.3), these equations include the well-known integrable equation



uxy = S(u) 1 − u2x (7.13)

where S is an arbitrary solution of the ordinary differential equation

S  + λ2 S = 0

and λ is some constant. Up to (complex) shifts and dilations in (7.13), there are
three distinct solutions of this ordinary differential equation:
• S = sin u,
• S = u,
• S = 1.
In the first case (7.13) is connected with the equation vxy = sin(v) by the differential
substitution v = arcsin(ux ) + u, and in the second case by the substitution v =
arcsin(ux ). In the third case (7.13) is a Liouville type equation (see Equation 4
in the list given in § 6) and can be reduced to the wave equation vxy = 0 by the
substitution v = arcsin(ux ).
In what follows, we present two more integrable equations (1.2). They are con-
nected by differential substitutions with the Tzitzéica equation.
Modified Tzitzéica equations. It turns out that the equation (3.24) considered
in § 3 is a degeneration of the apparently new equation

uxy = S(u)b(ux )b(uy ), (7.14)

which is integrable by the inverse scattering method. Here the functions b(ux) and
b(uy ) are determined by the formulae

(ux − b)(b + 2ux )2 = 1, (uy − b)(b + 2uy )2 = 1

and S is an arbitrary solution of the differential equation

S  − 2SS  − 4S 3 = 0. (7.15)

We note that (7.15) has the first integral

I = (S  − 2S 2 )2 (S  + S 2 ).

M. Pavlov informed us that the substitutions S = Q and Q2 = F (Q) reduce (7.15)
to the linear equation F  − 2F  − 8F = 0. Hence the function Z = exp (−2Q)
satisfies the equation Z 2 = c1 Z 3 + c2 , where the ci are constants.
1 1
The particular solution S = results in (3.24). The different solution S = −
u 2u
corresponds to the Liouville type equation
1
uxy = − b(ux )b(uy ).
2u
The latter equation is equivalent to one of the family of Equations 6 in § 6.
Exactly integrable hyperbolic equations of Liouville type 99

Equation (7.14) is connected with the Tzitzéica equation


vxy = c1 exp(v) + c2 exp(−2v)
by the differential substitution
1 1
v = − ln(ux − b) − ln(uy − b) + P (u),
2 2
where P (u) is determined from the differential equation
P 2 − 2SP  − 3S  − 2S 2 = 0.
The constants c1 and c2 satisfy the relation c21 c2 = −I/27.
Along with (7.14), the equation
uxy = S(u)b(ux ), (7.16)
where S  = 0, is connected with the Tzitzéica equation. Neglecting shifts and
dilations, there are two distinct cases:
• S = 3u,
• S = 1.
In the first case (see [46]), (7.16) is reduced to the Tzitzéica equation vxy = exp (v)+
exp (−2v) by the differential substitution
1
v = − ln(ux − b), (7.17)
2
which is invertible in the sense of [46].
In the second case, (7.16) is a Liouville type equation (see Class 1 in the list
in § 6) and can be reduced to the wave equation vxy = 0 by the substitution (7.17).
Conjecture. The equations presented in this subsection form an exhaustive (up
to the involution x ↔ y and complex transformations (6.1)) list of equations (1.2)
integrable by the inverse scattering method.
Finally, we note that the set of Laplace invariants Hi of a Liouville type equation
satisfies the system (2.22) with the termination condition Hr = H−s = 0. This
system is an open A-Toda lattice and hence is itself a Liouville type system.
The question arises as to how one can generalize this observation to the case of
sine-Gordon type equations. The most natural conjecture that the Laplace invari-
ants of such equations must satisfy periodically closed Toda lattices is false. It
would be of interest to describe the class of sine-Gordon type equations in terms of
their Laplace invariants.

Bibliography
[1] G. Darboux, “Sur les équations aux dérivées partielles du second ordre”, Ann. Sci. École
Norm. Sup. (7) (1870), 163–173.
[2] G. Darboux, Leçons sur la théorie générale des surfaces, vol. II, Hermann, Paris 1915.
[3] E. Goursat, Leçons sur l’intégration des équations aux dérivées partielles du second ordre à
deux variables indépendantes, vols. I, II, Hermann, Paris 1896, 1898.
[4] D. F. Egorov, “Partial differential equations of 2nd order in two independent variables.
General theory of integrals: characteristics”, Uchenye Zapiski Imperatorskogo Mosk. Univ.
15 (1899). (Russian)
100 A. V. Zhiber and V. V. Sokolov

[5] E. Gau, “Sur l’intégration des équations aux dérivées partielles du second ordre par la
méthode de M. Darboux”, J. Math. Pures Appl. 7 (1911), 123–240.
[6] E. Vessiot, “Sur les équations aux dérivées partielles du second ordre, F (x, y, p, q, r, s, t) = 0,
intégrables par la méthode de Darboux”, J. Math. Pures Appl. 18 (1939), 1–61.
[7] F. Tricomi, Equazioni differenziali, 2nd ed., Edizioni Scientifiche Einaudi, Torino 1953;
Russian transl., Lectures on partial differential equations, Inostr. Lit., Moscow 1957.
[8] V. E. Zakharov, C. V. Manakov, S. P. Novikov, and L. P. Pitaevskii, Theory of solitons. The
inverse scattering method, Nauka, Moscow 1980; English transl., Consultants Bureau, New
York–London 1984.
[9] L. A. Takhtadzhyan and L. D. Faddeev, Hamiltonian methods in the theory of solitons,
Nauka, Moscow 1986; English transl., Springer-Verlag, Berlin-New York 1986.
[10] R. Gosse, “La méthode de Darboux et les équations s = f (x, y, z, p, q)”, Mémorial de Sciences
Mathématiques, Fasc. 12, Paris 1926.
[11] E. Goursat, “Sur les intégrales de l’équation s = f (x, y, z, p, q)”, C. R. Acad. Sci. Paris 136
(1903), 1383–1384.
[12] É. Lainé, “Sur l’application de la méthode de Darboux aux equations s = f (x, y, z, p, q)”,
C. R. Acad. Sci. Paris 182 (1926), 1127–1128.
[13] É. Lainé, “Sur les équations s = f (x, y, z, p, q) intégrables par la méthode de Darboux”,
C. R. Acad. Sci. Paris 186 (1928), 209–210.
[14] D. H. Parsons, “The extension of Darboux’s method”, Mémorial de Sciences Mathématiques,
Fasc. 142, Gauthier-Villars, Paris 1960.
[15] M. Yu. Zvyagin, “Equations of second order reducible to zxy = 0 by a Bäcklund
transformation”, Dokl. Akad. Nauk SSSR 316 (1991), 36–40; English transl., Soviet Math.
Dokl. 43 (1991), 30–34.
[16] A. N. Leznov, V. G. Smirnov, and A. B. Shabat, “The group of internal symmetries and the
conditions of integrability of two-dimensional dynamical systems”, Teoret. Mat. Fiz. 51:1
(1982), 10–21; English transl., Theoret. and Math. Phys. 51 (1982), 322–330.
[17] I. M. Anderson and N. Kamran, “The variational bicomplex for second-order scalar
partial differential equations in the plane”, Preprint, Centre de Recherches Mathématiques,
Université de Montréal, Montréal 1994.
[18] I. M. Anderson and N. Kamran, “The variational bicomplex for hyperbolic second-order
scalar partial differential equations in the plane”, Duke Math. J. 87 (1997), 265–319.
[19] A. V. Zhiber, V. V. Sokolov, and S. Ya. Startsev, “On nonlinear Darboux-integrable
hyperbolic equations”, Dokl. Ross. Akad. Nauk 343 (1995), 746–748; English transl.,
Doklady Math. 52 (1996), 128–130.
[20] V. V. Sokolov and A. V. Zhiber, “On the Darboux integrable hyperbolic equation”, Phys.
Lett. A 208 (1995), 303–308.
[21] I. M. Anderson and M. Juras, “Generalized Laplace invariants and the method of Darboux”,
Duke Math. J. 89 (1997), 351–375.
[22] A. V. Zhiber and A. B. Shabat, “Klein–Gordon equations with a nontrivial group”, Dokl.
Akad. Nauk SSSR 247 (1979), 1103–1107; English transl, Soviet Phys. Dokl. 24 (1979),
607–609.
[23] A. V. Zhiber and A. B. Shabat, “Systems of equations ux = p(u, v), vy = q(u, v) possessing
symmetries”, Dokl. Akad. Nauk SSSR 277 (1984), 29–33; English transl., Soviet Math.
Dokl. 30 (1984), 23–26.
[24] A. B. Borisov and S. A. Zykov, “Dressing chain of discrete symmetries and breeding of
nonlinear equations”, Teoret. Mat. Fiz. 115:2 (1998), 199–214; English transl., Theoret.
and Math. Phys. 115 (1998), 530–541.
[25] A. V. Zhiber and V. V. Sokolov, “New example of a nonlinear hyperbolic equation possessing
integrals”, Teoret. Mat. Fiz. 120:1 (1999), 20–26; English transl., Theoret. and Math. Phys.
120 (1999), 834-839.
[26] J. Le Roux, “Extensions de la méthode de Laplace aux équations linéaires aux derivées
partielles d’ordre supérieur au second”, Bull. Soc. Math. France 27 (1899), 237–262.
[27] C. Athorne, “A Z2 × R3 Toda system”, Phys. Lett. A 206 (1995), 162–166.
Exactly integrable hyperbolic equations of Liouville type 101

[28] S. P. Tsarev, “Factoring linear partial differential operators and the Darboux method
for integrating nonlinear partial differential equations”, Teoret. Mat. Fiz. 122:1 (2000),
144–160; English transl., Theoret. and Math. Phys. 122 (2000), 121-133.
[29] A. V. Zhiber, “Quasilinear hyperbolic equations with an infinite-dimensional algebra of
symmetries”, Izv. Ross. Akad. Nauk Ser. Mat. 58:4 (1994), 33–54; English transl., Izv.
Math. 45 (1995), 33–54.
[30] A. B. Shabat and R. I. Yamilov, “Exponential systems of type 1 and Cartan matrices”,
Preprint, Bashkirskii Filial Akad. Nauk SSSR, Ufa 1981. (Russian)
[31] A. V. Zhiber, N. Kh. Ibragimov, and A. B. Shabat, “Equations of Liouville type”, Dokl. Akad.
Nauk SSSR 249 (1979), 26–29; English transl., Soviet Math. Dokl. 20 (1979), 607–609.
[32] A. N. Leznov, M. V. Savel’ev, and V. G. Smirnov, “General solutions of the two-dimensional
system of Volterra equations which realize the Bäcklund transformation for the Toda
lattice”, Teoret. Mat. Fiz. 47:2 (1981), 216–234; English transl., Theoret. and Math. Phys.
47 (1981), 417–422.
[33] V. V. Sokolov and A. B. Shabat, “Classification of integrable evolution equations”, Soviet
Sci. Rev. Sect. C. Math. Phys. Rev. 4 (1984), 221–280.
[34] A. S. Fokas, “Symmetries and integrability”, Stud. Appl. Math. 77 (1987), 253–299.
[35] A. V. Mikhailov, A. B. Shabat, and R. I. Yamilov, “A symmetric approach to the classification
of non-linear equations. Complete lists of integrable systems”, Uspekhi Mat. Nauk 42:4
(1987), 3–53; English transl., Russian Math. Surveys 42:4 (1987), 1–63.
[36] A. V. Mikhailov, A. B. Shabat, and V. V. Sokolov, “The symmetry approach to classification
of integrable equations”, What is integrability?, Springer, Berlin 1991, pp. 115–184.
[37] R. M. Miura, “Korteweg–de Vries equation and generalizations. I: A remarkable explicit
nonlinear transformation”, J. Math. Phys. 9 (1968), 1202–1204.
[38] S. I. Svinolupov, V. V. Sokolov, and R. I. Yamilov, “Bäcklund transformations for integrable
evolution equations”, Dokl. Akad. Nauk SSSR 271 (1983), 802–805; English transl., Soviet
Math. Dokl. 28 (1983), 165–168.
[39] V. V. Sokolov, “On the symmetries of evolution equations”, Uspekhi Mat. Nauk 43:5 (1988),
133–163; English transl., Russian Math. Surveys 43:5 (1988), 165–204.
d2 z
[40] T. Moutard, “Sur la construction des équations de la forme z1 dx dy
= λ(x, y), qui admettent
une intégrale générale explicite”, J. École Polytech. 45 (1878), 1–11.
[41] A. N. Leznov and M. V. Savel’ev, Group methods for the integration of nonlinear dynamical
systems, Nauka, Moscow 1985; English transl., Group-theoretical methods for integration of
nonlinear dynamical systems, Birkhäuser, Basel 1992.
[42] E. Goursat, “Sur une transformation de l’équation s2 = 4λ(x, y)pq”, Bull. Soc. Math. France
28 (1900), 1–6.
[43] P. J. Olver, Applications of Lie groups to differential equations, Oxford Univ., Math. Inst.,
Oxford 1980; Russian transl., Mir, Moscow 1989.
[44] A. B. Shabat, “Higher symmetries of two-dimensional lattices”, Phys. Lett. A 200 (1995),
121–133.
[45] S. Ya. Startsev, Laplace invariants for integrable hyperbolic equations and differential
substitutions of Miura type, Candidate (Ph.D) dissertation, Ufa Mat. Inst., Ufa 1997.
(Russian)
[46] V. V. Sokolov and S. I. Svinolupov, “On nonclassical invertible transformations of hyperbolic
equations”, European J. Appl. Math. 6 (1995), 145–156.

Institute of Mechanics,
Ufa Centre of the Russian Academy of Sciences
and
Centre for Non-linear Studies,
Institute for Theoretical Physics
Received 20/AUG/00

Typeset by AMS-TEX

You might also like