Hantjes Tensor

Math. Ann.
(2007) 339:61–99
DOI 10.1007/s00208-007-0106-2 Mathematische Annalen
Differential-geometric approach to the integrability

of hydrodynamic chains: the Haantjes tensor
E. V. Ferapontov · D. G. Marshall
Received: 30 August 2006 / Revised: 22 January 2007 / Published online: 12 April 2007
© Springer-Verlag 2007
Abstract The integrability of an m-component system of hydrodynamic type, ut =

V (u)ux , by the generalized hodograph method requires the diagonalizability of the
m × m matrix V (u). This condition is known to be equivalent to the vanishing of the
corresponding Haantjes tensor. We generalize this approach to hydrodynamic chains—
infinite-component systems of hydrodynamic type for which the ∞ × ∞ matrix V (u)
is ‘sufficiently sparse’. For such systems the Haantjes tensor is well-defined, and
the calculation of its components involves finite summations only. We illustrate our
approach by classifying broad classes of conservative and Hamiltonian hydrodynamic
chains with the zero Haantjes tensor. We prove that the vanishing of the Haantjes
tensor is a necessary condition for a hydrodynamic chain to possess an infinity of
semi-Hamiltonian hydrodynamic reductions, thus providing an easy-to-verify neces-
sary condition for the integrability.
Mathematics Subject Classification (2000) 35L40 · 35L65 · 37K10
1 Introduction
Hydrodynamic chains are quasilinear first order PDEs of the form
ut = V (u)ux (1)
E. V. Ferapontov (B) · D. G. Marshall

Department of Mathematical Sciences, Loughborough University,
Loughborough, Leicestershire LE11 3TU, UK
e-mail: E.V.Ferapontov@lboro.ac.uk
D. G. Marshall
e-mail: D.G.Marshall@lboro.ac.uk
123
62 E. V. Ferapontov, D. G. Marshall
where u = (u 1 , u 2 , . . .)t is an infinite-component column vector and V (u) is an ∞×∞

matrix. A classical example is the Benney chain (or Benney’s moment equations),
u nt = u n+1
x + (n − 1)u n−1 u 1x , (2)
n = 1, 2, . . ., which was derived in [2] from the equations for long nonlinear waves
on a free surface. It was observed later in [17,49] that the same system results from
a kinetic Vlasov equation. The system (2) has been thoroughly investigated in the
subsequent publications [17,25,26,31,36,46,49,50] where, in particular, its Hamilto-
nian and integrability aspects were uncovered. Hydrodynamic reductions of the chain
(2) were studied in [18–20,24,34]. Various deformations of Benney’s equations are
known. These include the modified Benney chain,
u nt = u n+1
x + u 1 u nx + (n − 1)u n u 1x , (3)
obtained in [28] as a quasiclassical limit of the modified KP hierarchy. Its two-param-

eter deformation,
u nt = u n+1
x + u 1 u nx + (a(n − 1) + b)u n u 1x , (4)
was constructed in [27] along with further examples of Hamiltonian integrable chains
possessing complete systems of commuting integrals. Another deformation scheme,
based on the R-matrix approach, was proposed in [3], see also [32,33]. The special-
ization of the chain (4) corresponding to a = 0, b = −1,
u nt = u n+1
x + u 1 u nx − u n u 1x , (5)
naturally appears in the theory of finite-gap solutions of integrable hierarchies of the

KdV type [1,35]. Reductions of the chain (5), both hydrodynamic and differential,
were extensively investigated in [37,44,45]. The case a = 1, b = 2 arises from
the kinetic model for rarefied bubbly flows (KR Khusnutdinova, SP Tsarev, private
communication), see also [47] for an alternative representation of this chain.
A broad class of new examples was found in [38], see also [4], based on the sym-
metry approach. These papers provide a classification of conservative chains of the
form
u 1t = u 2x , u 2t = g(u 1 , u 2 , u 3 )x , u 3t = h(u 1 , u 2 , u 3 , u 4 )x , . . . , (6)
which are embedded into a commutative hierarchy of Egorov’s type. It was observed
that the function g(u 1 , u 2 , u 3 ) uniquely determines all other equations of the chain
(6), as well as the whole associated hierarchy. Moreover, the function g(u 1 , u 2 , u 3 )
was shown to satisfy an over-determined involutive system of third order PDEs whose
general solution was expressed in terms of theta functions and solutions to the Chazy
equation.
123
Differential-geometric approach to the integrability of hydrodynamic chains 63
Our approach to the integrability of hydrodynamic chains is motivated by the theory

of finite-component systems of hydrodynamic type, that is, equations of the form (1)
where u = (u 1 , u 2 , . . . , u m )t is an m-component column vector and V (u) = v ij (u) is
an m × m matrix. Explicitly, one has
j
u it = v ij (u)u x , i, j = 1, ..., m; (7)
here and below we adopt the standard summation convention over the repeated upper
and lower indices. Such systems naturally occur in applications in gas dynamics, fluid
mechanics, chemical kinetics, Whitham averaging procedure, differential geometry
and topological field theory. We refer to [6,7,48] for a further discussion and refer-
ences. In what follows we consider the strictly hyperbolic case when the eigenvalues
of the matrix v ij , called the characteristic speeds of the system (7), are real and distinct.
It is known that many particularly important systems of the form (7) are diagonal-
izable, that is, reducible to the Riemann invariant form
Rti = λi (R)Rxi (8)
where the characteristic speeds λi (R) satisfy the so-called semi-Hamiltonian property
[48],

∂ j λi ∂k λi
∂k = ∂j , (9)
λ j − λi λ − λi
k
∂k = ∂/∂ R k , i = j = k. We emphasize that the semi-Hamiltonian property (9)

is usually automatically satisfied for diagonalizable systems of the ‘physical’ origin.
For instance, a conservative diagonalizable system is necessarily semi-Hamiltonian
(see Appendix 1 for more details). Such systems possess infinitely many conservation
laws and commuting flows of hydrodynamic type and can be linearized by the gener-
alized hodograph method [48]. Their analytic, differential-geometric and Hamiltonian
aspects are well-understood by now.
Remarkably, there exists an efficient tensor criterion of the diagonalizability which
does not require the actual computation of eigenvalues and eigenvectors of the matrix
v ij . Let us first calculate the Nijenhuis tensor of the matrix v ij ,
p p p p
N ijk = v j ∂u p vki − vk ∂u p v ij − v ip (∂u j vk − ∂u k v j ), (10)
and introduce the Haantjes tensor
p p p p
H ijk = N pr
i
v j vkr − N jr v ip vkr − Nr k v ip vrj + N jk vri vrp . (11)
For strictly hyperbolic systems the condition of diagonalizability is given by the fol-
lowing theorem.
123
Theorem 1 [22] A hydrodynamic type system with mutually distinct characteristic

speeds is diagonalizable if and only if the corresponding Haantjes tensor (11) is iden-
tically zero.
Since components of the Haantjes tensor can be calculated (using computer algebra)
in any coordinate system, this provides an efficient diagonalizability criterion.
Our main observation is that both tensors (10) and (11) make perfect sense for
infinite matrices which are ‘sufficiently sparse’. To be more rigorous, let us give the
following
Definition 1 An infinite matrix V (u) is said to belong to the class C (chain class) if
it satisfies the following two natural properties:
(a) each row of V (u) contains finitely many nonzero elements;
(b) each matrix element of V (u) depends on finitely many variables u i .
Notice that the chains (2–6) clearly belong to the class C. For matrices from the class
C all contractions in the expressions (10) and (11) reduce to finite summations so
that each particular component H ijk is a well-defined object which can be effectively
computed. Moreover, for a fixed value of an upper index i there exist only finitely
many non-zero components H ijk .
We propose the following
Definition 2 A hydrodynamic chain from the class C is said to be diagonalizable if
all components of the corresponding Haantjes tensor (11) are zero.
We point out that the chains (2–5) are diagonalizable in this sense. As we prove in
Sect. 5, the vanishing of the Haantjes tensor is a necessary (and in many particularly
important cases—sufficient) condition for a hydrodynamic chain to possess an infin-
ity of finite-component diagonalizable hydrodynamic reductions, thus justifying the
above definition. The main advantage of our approach is its ‘intrinsic’ character: it
does not require the knowledge of any ‘extrinsic’ objects such as commuting flows,
Hamiltonian structures, Lax pairs, etc.
The vanishing of the Haantjes tensor turns out to be an efficient classification cri-
terion. As an elementary example let us consider the chain
u nt = u n+1
x + u 1 u nx + cn u n u 1x
where cn = const. One can readily verify that the vanishing of the Haantjes tensor
implies the recurrence relation cn+2 = 2cn+1 − cn . Setting c1 = b, c2 = a + b we
recover the integrable chain (4).
Based on the same criterion, in Sect. 2 we classify diagonalizable chains of the type
1 = 0 are already sufficiently restrictive and
(6). It turns out that the conditions H jk
imply an over-determined system expressing all second order partial derivatives of the
function h in terms of g, see (14). The consistency conditions of these equations lead
to a closed-form involutive system expressing all third order partial derivatives of g
in terms of its lower order derivatives, see (15) (we have extensively used computer
algebra to calculate the Haantjes tensor and to verify the involutivity by calculating
123
the compatibility conditions). The requirement of the vanishing of other components

H ijk , i ≥ 2, imposes no additional constraints on h and g: these conditions reconstruct
the remaining equations of the chain (6). For instance, the conditions H jk2 = 0 specify
the right hand side of the fourth equation u t = . . ., etc.

4
The characterization of diagonalizable chains of a more general form,
u 1t = f (u 1 , u 2 )x , u 2t = g(u 1 , u 2 , u 3 )x , u 3t = h(u 1 , u 2 , u 3 , u 4 )x , . . . , (12)
etc, is proposed in Sect. 3. As in the previous example, the conditions H jk1 = 0 lead
to expressions for all second order partial derivatives of h in terms of g and f . The
consistency conditions of these equations result in an involutive system expressing all
third order partial derivatives of g and f in terms of lower order derivatives thereof.
In Sect. 4 we classify diagonalizable Hamiltonian chains of the form

d d t ∂h
ut = B + B , (13)
dx dx ∂u
here B i j = (i − 1)u i+ j−2 and h(u 1 , u 2 , u 3 ) is a Hamiltonian density. The Benney

chain (2) corresponds to h = (u 3 + (u 1 )2 )/2, see [26]. We have found a broad family
of new integrable Hamiltonian densities, thus extending the results of [27].
In Sect. 5 we prove that the condition of diagonalizability is necessary for the exis-
tence of ‘sufficiently many’ hydrodynamic reductions. Recall that an m-component
hydrodynamic reduction of an infinite chain is specified by parametric equations
u 1 = u 1 (R 1 , . . . , R m ), u 2 = u 2 (R 1 , . . . , R m ), u 3 = u 3 (R 1 , . . . , R m ), . . . ,
etc, where the Riemann invariants R 1 , . . . , R m solve a diagonal system (8) whose
characteristic speeds satisfy the semi-Hamiltonian property (9). It is required that all
equations of the chain are satisfied identically modulo (8). Thus, an infinite chain
reduces to a system with finitely many dependent variables. It was demonstrated in
[19] that the Benney chain (2) possesses infinitely many m-component reductions
of this type parametrized by m arbitrary functions of a single variable. The same
is true for other chains (2–6). Based on these and other examples we propose the
following:
Definition 3 A hydrodynamic chain from the class C is said to be integrable if, for
any m, it possesses infinitely many m-component semi-Hamiltonian reductions para-
metrised by m arbitrary functions of a single variable.
Remark It was observed in [30], (MV Pavlov, private communication, 2005) that
Hamiltonian chains (13) possess m-component reductions for any Hamiltonian den-
sity h, even in the non-integrable case. The crucial difference is that for integrable
chains these reductions are semi-Hamiltonian and depend on m arbitrary functions of
a single variable, while in the non-integrable situation for any m there exists a unique
m-component reduction which is not diagonalizable.
123
In Sect. 5 we prove our main result:
Theorem 2 The vanishing of the Haantjes tensor H is a necessary condition for the
integrability of hydrodynamic chains from the class C.
If the spectrum of the infinite matrix V is simple, that is, for a generic λ there exists
a unique formal eigenvector ξ such that V ξ = λξ (we point out that all of the above
examples satisfy this property), one has the following stronger result:
Theorem 3 In the simple spectrum case the vanishing of the Haantjes tensor H is
necessary and sufficient for the existence of two-component reductions parametrized
by two arbitrary functions of a single variable.
Theorem 2 provides an easy-to-verify necessary condition for testing the integra-

bility of hydrodynamic chains. We emphasize that the vanishing of the Haantjes tensor
is not sufficient for the integrability in general: one can construct examples of diag-
onalizable chains which possess infinitely many diagonal reductions none of which
are semi-Hamiltonian (see Sect. 5). To eliminate these pathological cases let us recall
that for finite-component systems (7) there exists a tensor object which is responsible
for the semi-Hamiltonian property [39]. This is a (1, 3)-tensor Pkis j (see Appendix 1
for explicit formulas in terms of the matrix v ij ). Similarly to the Haantjes tensor H ,
the tensor P is well-defined for hydrodynamic chains from the class C. We conclude
this introduction by formulating the following:
Conjecture The vanishing of both tensors H and P is necessary and sufficient for the
integrability of hydrodynamic chains from the class C.
The necessity of this conjecture (that is, the statement that the integrability implies
the vanishing of both H and P) is a relatively simple fact, see Sect. 5 for a proof.
The sufficiency is a far more delicate property which we were not able to establish
in general. We point out that the vanishing of H alone (in fact, the vanishing of the
very few first components of H ), is already sufficiently restrictive and implies the
integrability in many particularly important cases (e.g. for conservative chains, Ham-
iltonian chains, etc). Recall that, for finite-component systems of hydrodynamic type,
our conjecture is a well-known result: any diagonalizable semi-Hamiltonian system
possesses infinitely many conservation laws and commuting flows of hydrodynamic
type, and can be solved by the generalized hodograph method [48].
2 Classification of diagonalizable chains of the type (6)
The structure of Eqs. (6) implies that the only nonzero components H ijk of the Haantjes
tensor are the ones with j ≤ i + 4, k ≤ i + 4. Taking into account the skew-sym-
metry of the Haantjes tensor in its lower indices, this leaves ten essentially different
components of the type H jk 1 . Equating them to zero we obtain the expressions for all
123
of the ten second order partial derivatives of h(u 1 , u 2 , u 3 , u 4 ):
2g1 g12 − g2 g11 + 2h 1 g13

h 11 = ,
g3
g22 g1 + g11 + g13 h 2 + g23 h 1
h 12 = ,
g3
g2 g22 + 2g12 + 2g23 h 2
h 22 = ,
g3
g13 (h 3 − g2 ) + g23 g1 + g12 g3 + g33 h 1
h 13 = , (14)
g3
g13 + h 3 g23 + h 2 g33
h 23 = g22 + ,
g3
g33 (g2 − 2h 3 )
h 33 = 2g23 − ,
g3
h 4 g13 h 4 g23 h 4 g33
h 14 = , h 24 = , h 34 = , h 44 = 0.
g3 g3 g3
Notice that these equations can be compactly written as
2 1
d2h = (dhdg3 + dg1 du 2 + dg2 dg − g2 d 2 g);
g3 2
here both sides of the equality are understood as symmetric two-forms, and dg, dh,
d 2 g, d 2 h denote the first and second symmetric differentials of g and h. The consis-
tency conditions for the Eqs. (14) lead to closed-form expressions for all third order
partial derivatives of the function g(u 1 , u 2 , u 3 ) in terms of its lower order derivatives:
2
2g33 2g13 g33 2g23 g33
g333 = , g133 = , g233 = ,
g3 g3 g3
2g 2 2g13 g23 2g 2
g113 = 13 , g123 = , g223 = 23 ,
g3 g3 g3
2
g222 = 2 g2 g232
+ g23 (g3 g22 + 2g13 ) − g33 (g2 g22 + 2g12 ) ,
g3
2
g122 = 2 g1 g232
+ g13 (g3 g22 + g13 ) − g33 (g1 g22 + g11 ) , (15)
g3
2
g112 = 2 (g33 (g2 g11 − 2g1 g12 ) − g13 (g2 g13 − 2g3 g12 ) − g23 (g3 g11 − 2g1 g13 )) ,
g3
2
g111 = 2 (g1 + g22 )g132
+ g12 g23
2
+ g32 (g12
2
− g11 g22 ) − g22 g33 g12
g3
+g13 g3 (g11 + 2(g1 g22 − g2 g12 )) + 2g23 (g2 (g3 g11 − g1 g13 ) − g1 g3 g12 )

− g33 ((g1 + g22 )g11 − 2g1 g2 g12 ) .
123
This system is in involution and its general solution depends on ten integration con-
stants, indeed, the values of g and its partial derivatives up to the second order can be
prescribed arbitrarily at any fixed point u 10 , u 20 , u 30 . The system (15) was first derived
in [38] from the requirement that the chain (6) is embedded into a hierarchy of com-
muting hydrodynamic chains of Egorov’s type. Exactly the same equations for g were
obtained in [14] by applying the method of hydrodynamic reductions to the (2+1)-
dimensional PDE
u tt = g(u x x , u xt , u x y ) (16)
which is naturally associated with the chain (6); here the function g is the same as in
(6, 15). Thus, for hydrodynamic chains of the type (6) the condition of diagonalizability
is necessary and sufficient for the integrability.
One can show that the vanishing of other components of the Haantjes tensor does
not impose any additional constraints on the derivatives of g and h. Thus, writing
the fourth equation of the chain (6) in the form u 4t = s(u 1 , u 2 , u 3 , u 4 , u 5 )x and set-
2 = 0, one obtains the expressions for all second order partial derivatives
ting H jk
of s in terms of h and g, which are analogous to (14). The consistency conditions
are satisfied identically modulo (14, 15). Similarly, the condition H jk 3 = 0 speci-
fies the right hand side of the fifth equation of the chain, etc. Although we know no
direct way to demonstrate the non-obstructedness of this recursive procedure in gen-
eral, there exists an alternative direct approach to the reconstruction of a chain from
the function g(u 1 , u 2 , u 3 ). To illustrate this procedure we consider a simple example
g = u 3 − 21 (u 1 )2 which automatically satisfies (15). The corresponding h, as specified
by (14), is given by h = µ + αu 1 + βu 2 + γ u 3 + δu 4 − u 1 u 2 , which can be reduced to
a canonical form h = u 4 − u 1 u 2 by redefining u 4 appropriately (this transformation
freedom allows one to absorb arbitrary integration constants arising at each step of
the construction). Thus, the first three equations of the chain are
1
u 1t = u 2x , u 2t = (u 3 − (u 1 )2 )x , u 3t = (u 4 − u 1 u 2 )x , . . . , (17)
2
etc. Equations (17) are nothing but the first three equations in the conservative repre-
sentation of the Benney chain (2); notice that the variables u i in (2) and (17) are not
the same: u i in (17) are conserved quantities of the chain (2). The reconstruction of
the remaining equations of the chain consists of three steps.
(i) One introduces the corresponding PDE (16): in our case this will be a potential
form of the dispersionless KP equation, u tt = u x y − 21 u 2x x . It was demonstrated
in [14] that the general PDE (16) is integrable by the method of hydrodynamic
reductions if and only if the function g satisfies the relations (15).
(ii) One constructs a dispersionless Lax pair for the PDE (16), in our example it
takes the form

1 2 1 3
pt = p + uxx , py = p + u x x p + u xt ;
2 x 3 x
123
the consistency conditions pt y = p yt are satisfied identically modulo the dis-

persionless KP equation. The existence of such Lax pairs was established in
[38,14] for any equation (16) provided g satisfies the compatibility conditions
(15).
(iii) One looks for p as an expansion in the auxiliary parameter λ,
u1 u2 u3
p =λ− − 2 − 3 − ··· ;
λ λ λ
the substitution of this ansatz into the first equation of the Lax pair implies an
infinite hydrodynamic chain for the variables u i . The first three equations of this
chain identically coincide with (17). The substitution into the second equation of
the Lax pair produces a commuting chain (one has to set u 1 = u x x , u 2 = u xt ).
Both chains possess infinitely many hydrodynamic reductions since this is the
case for the generating Eq. (16). Thus, the Haantjes tensor will automatically
vanish (as demonstrated in Sect. 5).
This procedure has been successfully implemented in [40,41] for various classes of
solutions of the system (15), namely, cases (18) and (19) of the classification presented
below (according to [38], the case (19) is reciprocally related to the case (18) and,
therefore, does not require a special treatment). The work on (20) and (21) is currently
in progress (Pavlov, M.V.: private communication, 2005).
2.1 Integration of the system (15)
To explicitly calculate g(u 1 , u 2 , u 3 ) we will follow [38]. The main observation is

that the first six equations in (15) imply that the function 1/g3 is linear, 1/g3 =
α + βu 1 + γ u 2 + δu 3 . If δ = 0 then, up to a linear change of variables, one can
assume that 1/g3 = u 3 . Similarly, if δ = 0, γ = 0, one can set 1/g3 = u 2 . If
δ = γ = 0, β = 0 one has 1/g3 = u 1 . The last possibility is 1/g3 = 1. Thus, we
have four cases to consider:
u3 u3
g = u 3 + p(u 1 , u 2 ), g = + p(u ,
1 2
u ), g = + p(u 1 , u 2 ),
u1 u2
g = ln u 3 + p(u 1 , u 2 );
here the function p(u 1 , u 2 ) can be recovered after the substitution into the remaining
four Eqs. (15). In each of these cases the resulting equations for p(u 1 , u 2 ) integrate
explicitly, see [38], leading to the four essentially different canonical forms:
1
(Au 2 + 2Bu 1 )2 + Ce−Au ,
1
g = u3 + (18)
4A

u3 1 A B B2
− Ce−A/u ,
1
g= 1+ − (u 2 )2 + 1 2 u 2 − (19)
u u 1 4(u )
1 2 (u ) A(u 1 )2
123
u3 1
g= + η(u 1 )(u 2 )2 , (20)
u2 6
1
g = ln u − ln σ u , u −
3 1 2
η(u 1 )du 1 . (21)
4
Here η(u 1 ) is a solution to the Chazy equation [5],
η + 2ηη − 3η = 0,

2
(22)
and σ (u 1 , u 2 ) is an elliptic sigma function in the variable u 2 whose dependence on

u 1 is governed by the Chazy equation (see Case IV below). Details of the derivation
of canonical forms (18–21) can be summarized as follows.
Case I. Substituting the ansatz g = u 3 + p(u 1 , u 2 ) into (15) one arrives at the equations
p111 = 2( p12
2
− p11 p22 ),
p112 = p122 = p222 = 0.
The last three equations imply
1
p= A(u 2 )2 + (Bu 1 + D)u 2 + q(u 1 ),
4
and the substitution into the first equation results in the linear ODE q + Aq = 2B 2 .
Up to a transformation of the form u 3 → u 3 + αu 2 + βu 1 + γ this leads to (18).
Case II. Substituting the ansatz g = u 3 /u 1 + p(u 1 , u 2 ) into (15) one arrives at the
equations
2 2
p111 = 2( p12
2
− p11 p22 ) + ( p1 + p22 ) − 1 ( p11 + 2 p1 p22 − 2 p2 p12 ),
(u )
1 2 u
2 4
p112 = − p2 − 1 p12 ,
(u 1 )2 u
2 2
p122 = 1 2 − 1 p22 , p222 = 0.
(u ) u

1 A D B B2
p= − (u ) +
2 2
+ 1 2 u2 − + q(u 1 ),
u 1 4(u 1 )2 u 1 (u ) A(u 1 )2
and the substitution into the first equation results in the linear ODE (u 1 )3 q +u 1 (6u 1 −
A)q + (6u 1 − 2 A)q = 0 whose basis of solutions consists of 1, 1/u 1 and e−A/u .
1
Up to a transformation of the form u 3 → u 3 + αu 2 + βu 1 + γ this implies (19). It was

observed in [38] that the cases I and II are reciprocally related: under the change from
x, t to the new independent variables X, T defined as d X = u 1 d x + u 2 dt, T = t, and
2 3
the introduction of the new dependent variables U 1 = u11 , U 2 = − uu 1 , U 3 = − uu 1 ,
123
etc, the chains from the case I transform to the chains from the case II, and vice versa.
On the level of the corresponding equations this means that the change of variables
U 1 = u11 , U 2 = − uu 1 , P = (uu 1) − p, G = (uu 1) − g transforms the equations for
2 2 2 2 2
p from the Case I to the equations for p from the Case II. Equivalently, (18) goes to
(19).
Case III. Substituting the ansatz g = u 3 /u 2 + p(u 1 , u 2 ) into (15) one arrives at the
equations
2 4
p111 = p12 + 2( p12
2
− p11 p22 ) + 2 ( p1 p12 − p2 p11 ),
(u )
2 2 u
2 2
p112 = 2 p11 , p122 = 2 2 p1 ,
u (u )
2 2
p222 = 2 2 p2 − 2 p22 .
(u ) u
B + Cu 1 1
p= A+ 2
+ η(u 1 )(u 2 )2 ,
u 6
and the substitution into the first relation results in the Chazy equation (22) for η.
Elimination of the constants A, B, C leads to (20).
Case IV. Substituting the ansatz g = ln u 3 + p(u 1 , u 2 ) into (15) one arrives at the
equations
p111 = 2( p12
2
− p11 p22 ) + 2( p1 + p22 ) p11 − 4 p1 p2 p12 + 2 p12 p22 ,
p112 = 4 p1 p12 − 2 p2 p11 ,
p122 = 2 p11 + 2 p1 p22 ,
p222 = 4 p12 + 2 p2 p22 .
The general solution of the fourth equation can be represented in the form
1
p = − ln σ u , u −
1 2
η(u 1 )du 1 ;
4
here σ solves the heat equation 4σ1 = σ22 and η is a function of u 1 . It is convenient
to introduce the new variable v(u 1 , u 2 ) by the formula v = −(ln σ )22 . Taking into
account the heat equation for σ one has
1 1 1
v1 = v22 − v 2 + (ln σ )2 v2 .
4 2 2
Rewritten in terms of v, the third equation for p implies
v22 = 6v 2 − 4vη − 4η , (23)
123
≡ d/du 1 , the second equation is satisfied identically and the first takes the form
8
v22 = 4v 3 − 4v 2 η − 8vη − η . (24)
3
This shows that v is a shift of the Weierstrass elliptic function in the variable u 2 . Since
v = −(ln σ )22 , the function σ is the corresponding theta function. Notice that (23) can
be obtained as a result of differentiation of (24) by u 2 . Thus, we have two equations
for v:
1
v1 = v 2 − vη − η + (ln σ )2 v2 ,
2
8
v2 = 4v − 4v η − 8vη − η .
2 3 2
3
The condition of their consistency leads to the Chazy equation for η.
3 Classification of diagonalizable chains of the type (12)
In this section we sketch the classification of diagonalizable chains of the type (12). The
condition H jk1 = 0 implies the following formulae for second order partial derivatives
of h:
h 4 g13 h 4 g23 h 4 g33

h 14 = , h 24 = , h 34 = , h 44 = 0,
g3 g3 g3
f 22 g1 g13 (h 3 − g2 ) + g23 g1 + g12 g3 + g33 h 1
h 13 = − + ,
f2 g3
( f 1 − g2 ) f 22 (h 3 − f 1 )g23 + f 2 g13 + h 2 g33
h 23 = g22 − f 12 + + ,
f2 g3
f 22 g3 g33 ( f 1 + g2 − 2h 3 )
h 33 = 2g23 − − , (25)
f2 g3
f 22 ( f 12 − 2 f 1 g2 + g22 + f 2 g1 )
h 22 = −
g3 f 2
(g2 − f 1 )(g22 − 2 f 12 ) + f 2 (2g12 − f 11 ) + 2g23 h 2
+ ,
g3
f 22 g1 ( f 1 − g2 ) g22 g1 + f 2 g11 + g13 h 2 + g23 h 1 − 2g1 f 12
h 12 = + ,
g3 f 2 g3
g12 f 22 2g1 g12 + ( f 1 − g2 )g11 + 2h 1 g13 − g1 f 11
h 11 = − + .
g3 f 2 g3
By calculating the consistency conditions for the above equations we obtain the expres-
sions for all third order partial derivatives of g and f .
123
Equations for g:
2
2g33 2g13 g33 2g23 g33
g333 = , g133 = , g233 = ,
g3 g3 g3
2
2g13 2g 2
2g13 g23
g113 = , g123 = , g223 = 23 ,
g3 g3 g3
f 11 (−g33 g2 g1 +g3 (g23 g1 +g2 g13 −g3 g12 ))+2 f 12 (g13

2 −g g )+ f f (g g −g g )
33 11 1 11 33 1 3 13
g111 =
f 2 g32
2 f 1 ((−g33 g1 + g3 g13 )g12 + g23 (g1 g13 − g3 g11 ) + g2 (−g13
2 + g g )) + g 2 g 2
33 11
+2 23 1
f 2 g32
g22 (−g12 g33 +2g3 g1 g13 −g32 g11 )+g13
2 (g 2 + f g )+g (2g g g −2g g g +g 2 g )
2 1 12 33 2 1 2 3 13 3 12
+2 2
f 2 g32
g11 (−g33 g22 − f 2 g33 g1 + f 2 g3 g13 ) − 2g23 (g1 g3 g12 + g2 (g1 g13 − g3 g11 ))
+2
f 2 g32
f 12 (g33 g12 + g3 (−2g1 g13 + g3 g11 ))
+2 , (26)
f 2 g32
(g33 g12 + g3 (g3 g11 − 2g1 g13 )) f 22 + f 2 ( f 11 (g33 g1 − g3 g13 ) − 2g13 (g2 g13 − 2g3 g12 ))
g112 =
f 2 g32
g33 (2g1 g12 − g2 g11 ) + g23 (g3 g11 − 2g1 g13 ) + f 1 (g33 g11 − g13
2 )
−2 ,
g32
(g1 g2 g33 + (g3 g13 − g1 g33 ) f 1 + (g3 g12 − g23 g1 − g2 g13 )g3 )
g122 = f 22
f 2 g32
2 g − g g g + g g g + f g 2 + (g g − g g ) f − f g g )
2(g23 1 33 22 1 3 22 13 2 13 33 1 3 13 12 2 33 11
+ ,
g32
((( f 1 −2g2 )g33 +2g3 g23 ) f 1 + (g22 + f 2 g1 )g33 + (g3 g22 − 2g2 g23 − f 2 g13 )g3 ) f 22
g222 =
f 2 g32
2 −g g )+ f (g g −g g )+g g g +g (g 2 −g g )
− f 1 ( f 12 g33 +g23 33 22 12 33 2 3 23 3 23 22 2 23 22 33
+2
g32
f 2 (g33 ( f 11 − 4g12 ) + 4g23 g13 )
+ .
g32
Equations for f :
f 11 (g33 ( f 12 +g22 + f 2 g1 )−2g2 g3 g23 +g32 g22 − f 2 g3 g13 )−2 f 1 f 11 (g33 g2 −g3 g23 )
f 111 = −
f 2 g32
g 2 f 11 + 2(g1 g2 g33 + f 1 (−g33 g1 + g3 g13 ) + g3 (−g23 g1 − g2 g13 + g3 g12 ))
+ f 12 3
f 2 g32
f 22 (g33 g12 − 2g1 g3 g13 + g32 g11 )
− , (27)
f 2 g32
123
f 22 f 11 g32 − f 2 ( f 1 f 11 g33 + f 11 (g3 g23 − g33 g2 ) + 2 f 12 (g33 g1 − g3 g13 ))

f 112 = ,
f 2 g32
− f 22 f 11 g33 + f 22 ( f 12 g32 + f 2 (−g33 g1 + g3 g13 ))
f 122 = ,
f 2 g32
2 g 2 − 2 f 2 f g + f f (( f − g )g + g g )
f 22
f 222 = 3 2 12 33 2 22 1 2 33 3 23
.
f 2 g32
We have verified that the system (26, 27) is in involution. We claim that the functions
f and g contain all the necessary information about conservative chains (12). In par-
ticular, Eqs. (25) allow one to reconstruct the function h and, hence, the right hand side
2 = 0 reconstructs the
of the third equation of the chain. Similarly, the requirement H jk
fourth equation (i.e., provides an involutive second order system for the right hand side
of the fourth equation), etc. Although we have verified directly that our procedure is
non-obstructed up to order 5, we could not establish this property in general. Possible
approaches to this problem could be
– developing a cohomological approach to the reconstruction procedure in the spirit
of [8];
– establishing a Hamiltonian formulation for general integrable chains of the type
(12);
– establishing a link of integrable chains of the type (12) to (2 + 1)-dimensional
integrable PDEs, as explained in Sect. 2. Although there is a general belief that
one can construct a one-to-one correspondence between integrable hydrodynamic
chains and integrable (2 + 1)-dimensional quasi-linear PDEs (that is, any integra-
ble PDE can be ‘decoupled’ into a pair of commuting chains), there exists yet no
rigorous proof of this statement.
3.1 Integration of the Eqs. (26), (27)
Notice that the first six equations in (26) are exactly the same as in Sect. 2. Thus, there
are four essentially different cases to consider.
Case I: g = u 3 + p(u 1 , u 2 ). Substituting this ansatz into (26, 27) one arrives at the
following relations.
equations for p:
p12 (2 p12 − f 11 ) + 2 p11 ( f 12 − p22 )

p111 = ,
f2
f 22 p11 f 22 p12 f 22 p22
p112 = , p122 = , p222 = ;
f2 f2 f2
equations for f :
f 11 ( f 12 − p22 ) + 2 f 12 p12 − f 22 p11

f 111 = ,
f2
f 22 f 11 f 22 f 12 f2
f 112 = , f 122 = , f 222 = 22 .
f2 f2 f2
123
The last three equations for f and the last three equations for p lead, up to elementary
changes of variables, to the two possibilities.
Subcase I1 :
2 2
f = s(u 1 )eu , p = q(u 1 )eu .
The substitution of these expressions into the remaining equations for p111 and f 111
leads to a system of coupled ODEs for s(u 1 ) and q(u 1 ):
2((q )2 − qq ) − q s + 2q s
q = ,
s
s s + 2s q − qs − q s
s = .
s
Setting q = −s , the second equation will be satisfied identically while the first implies
a fourth order ODE s s + 3(s )2 − 4s s = 0 whose general solution is an ellip-
tic sigma-function: s = σ (u 1 ), here (ln σ ) = −℘, (℘ )2 = 4℘ 3 − c (notice that
g2 = 0, g3 = c). Thus, as a particular case we have
p = −σ (u 1 )eu .
2 2
f = σ (u 1 )eu ,
Subcase I2 :
1
f = (au 1 + b)u 2 + s(u 1 ), p= A(u 2 )2 + Bu 1 u 2 + q(u 1 ),
2
a, b, A, B=const. The substitution of these expressions into the remaining equations

for p111 and f 111 leads to linear ODEs for s(u 1 ) and q(u 1 ):
2B 2 − Bs + 2(a − A)q
q = ,
au 1 + b
(a − A)s + 2a B
s = .
au 1 + b
These equations are straightforward to solve. One needs to consider two different
cases: a = 0, b = 1 and a = 1, b = 0. If a = 0, b = 1 then, up to unessential
integration constants, we have
B2 1 2 B
s = αe−Au , q = (u ) − α e−Au + βe−2 Au .
1 1 1
2A A
The case a = 1, b = 0 leads to
B B 2 (A − 2) 1 2 αB
s= (u 1 )2 +α(u 1 )3−A , q = (u ) + (u 1 )3−A +β(u 1 )2(2−A) .
A−1 2(A − 1)2 1− A
123
Case II: g = u 3 /u 1 + p(u 1 , u 2 ). Substituting this ansatz into (26), (27) one arrives
at the following set of relations.
equations for p:
p22 + f 12 + f 2 p1 − 4 p2 f 1 ( f 1 − p2 )( f 11 − 4 p12 ) + 4 p1 ( f 12 − p22 )

p111 = 2 +
f 2 (u 1 )2 u 1 f2
p12 (2 p12 − f 11 ) + 2 p11 ( f 12 − p22 ) p11
+ −2 1 ,
f2 u
2 f 11 − 4 p12 2 p1 f 22 f 22 p11
p112 = 1 2 ( f 1 − p2 ) + + 1 + ,
(u ) u 1 u f2 f2
f2 2 f 22 ( p2 − f 1 ) f 22 p12
p122 = 2 1 2 + 1 ( f 12 − p22 ) + + ,
(u ) u 1
u f2 f2
f 22 p22 f 22
p222 = + 1;
f2 u
equations for f :

f 11 f2 f 12 f 22
f 111 = f 12 − p22 − 1 + 1
(2 p2 − 2 f 1 + p12 u 1 ) − (2 p1 + p11 u 1 ),
f2 u f2 u f2 u 1
2
f 22
f 11 f 22 f 12 f 12 f 22 f 22
f 112 = −2 1 , f 122 = − 1, f 222 = .
f2 u f2 u f2
The cases I and II are reciprocally related (we thank Maxim Pavlov for pointing out this
equivalence): under the change from x, t to the new independent variables X, T defined
as d X = u 1 d x + f dt, T = t, and the introduction of the new dependent variables
2 3
U 1 = u11 , U 2 = − uu 1 , U 3 = − uu 1 , etc, the chains from the case I transform to the
chains from the case II, and vice versa. On the level of the corresponding equations this
2 2
means that the change of variables U 1 = u11 , U 2 = − uu 1 , F = − uf1 , P = uu 1 f − p
transforms the equations for p, f from the Case I to the equations for P, F from the
Case II. Thus, we will not discuss this case in any more detail here.
Case III: g = u 3 /u 2 + p(u 1 , u 2 ). Substituting this ansatz into (26), (27) one arrives
at the following set of relations.
equations for p:
2 p12 p1 (4 p12 − f 11 ) + 4 p11 ( f 1 − p2 )

p111 = +
f 2 (u )
2 2 u 2 f2
p12 (2 p12 − f 11 ) + 2 p11 ( f 12 − p22 )
+ ,
f2
f 22 p11 p11 2 f 2 p1 + u 2 f 22 ( p1 + u 2 p12 )
p112 = + 2 2 , p122 = ,
f2 u f 2 (u 2 )2
f 22 ( p2 − f 1 ) − f 2 p22 p22 f 22 f 12 2
p222 =2 + + 2 2 − 2 2 ( f 1 − p2 );
u 2 f2 f2 u (u )
123
equations for f :
f 11 f 12 2 f 22 p11
f 111 = (2( f 1 − p2 ) − p22 u 2 ) + (u ( f 11 + 2 p12 ) + 2 p1 ) − ,
f2 u 2 f2 u 2 f2
f 22 f 11 f 11 f 22 f 12 f2 f 22
f 112 = + 2 , f 122 = , f 222 = 22 − 2 .
f2 u f2 f2 u
The last three equations for f and the last three equations for p lead to the two essen-
tially different possibilities:
Subcase III1 :
f = s(u 1 )(u 2 )k , p = q(u 1 )(u 2 )k+1 ,
k=const. The substitution of these expressions into the remaining equations for p111
and f 111 leads to the coupled system of ODEs for s(u 1 ) and q(u 1 ):
k+2
q = (2(k + 2)(q )2 − 2(k + 1)qq − q s + 2q s ),
ks
k + 2
s = (s s − k(k + 1)s q + 2ks q ) − (k − 1)q .
ks
Notice that under the substitution s = k = 1 the equation for q reduces to the Chazy
equation (22) for q(u 1 ) = 16 η(u 1 ).
Subcase III2 :
f = (au 1 + b) ln u 2 + s(u 1 ), p = q(u 1 )u 2 + a(u 2 ln u 2 − u 2 ),
a, b=const. The substitution of these expressions into the remaining equations for p111
and f 111 leads to the coupled system of ODEs for s(u 1 ) and q(u 1 ):
8(q )2 − 4qq − 2q s + 4q s
q = ,
au 1 + b
s (s − q) + 2aq
s =2 + q .
au 1 + b
Case IV: g = ln u 3 + p(u 1 , u 2 ). Substituting this ansatz into (26, 27) one obtains
equations for p:
p22 (2 p12 −2 p11 )+ p11 ( p2 (2 p2 −4 f 1 )+2( f 12 + f 12 + f 2 p1 ))+ p12 (2 p12 − f 11 +4 p1 ( f 1 − p2 ))

p111 =
f2
p1 (2 p1 f 12 + ( f 1 − p2 ) f 11 )
− ,
f2
f 22 ( p11 − p12 )
p112 = + p1 (4 p12 − f 11 ) + 2 p11 ( f 1 − p2 ),
f2
123
f 22
p122 = ( p1 ( f 1 − p2 ) + p12 ) + 2 p1 ( p22 − f 12 ) + 2 f 2 p11 ,
f2
f 22
p222 = ( p22 + f 1 (2 p2 − f 1 ) − p22 ) + 2( p22 − f 12 )( p2 − f 1 ) − f 22 p1 + f 2 (4 p12 − f 11 );
f2
equations for f :
f 11 2
f 111 = ( f + p22 + f 2 p1 − p22 − 2 f 1 p2 + f 12 )
f2 1
f 12 f 22 2
+2 ( p1 ( f 1 − p2 ) + p12 ) + ( p − p11 ),
f2 f2 1
f 22 f 11 f 22 f 12
f 112 = + f 11 ( f 1 − p2 ) + 2 f 12 p1 , f 122 = + f 2 f 11 + f 22 p1 ,
f2 f2
2
f 22
f 222 = + 2 f 2 f 12 + f 22 ( p2 − f 1 ).
f2
Although this system is in involution and reduces to the corresponding Case 4 in

Sect. 2 under the substitution f = u 2 , we were not able to integrate it in general. Let
us just mention that the last three relations for f imply the Monge–Ampere equation
f 11 f 22 − f 12
2 = a(u 1 ) f 2 . This suggests a separable ansatz f = s(u 1 )r (u 2 ). A simple
2
analysis leads to the two possibilities:
1 1
(i) f = s(u 1 )(u 2 )k , p = 1−k s (u )(u 2 )k+1 − ln u 2 , where s satisfies the equation
ss − k−1k
(s )2 = 0. Without any loss of generality one can take s = (u 1 )1−k .
(ii) f = s(u 1 )eu , p = −s (u 1 )eu , where s satisfies the equation ss − (s )2 = 0.
2 2
1
Up to a linear transformation of u 1 one has s = eu .
4 Classification of diagonalizable Hamiltonain chains
It was observed in [25,26] that the Benney chain (2) can be represented in the Hamil-
tonian form (13),

d d t ∂h
ut = B + B ,
dx dx ∂u
where B i j = (i − 1)u i+ j−2 , and h = (u 3 + (u 1 )2 )/2 is the Hamiltonian density.

Further integrable examples can be constructed by looking at Hamiltonian densities
in the form h = u 3 + p(u 1 , u 2 ), and imposing the constraint H jk
1 = 0. This implies
the relations
p111 (2 + u 1 p22 ) = p11 p22 − p12

2
, p112 = p122 = p222 = 0,
123
which, up to a natural equivalence h → αh + au 1 + bu 2 + c, lead to the Hamiltonian

densities
h = u 3 + α(u 1 )2 + βu 1 u 2 + γ (u 2 )2 + δ(u 1 )3 (28)
where the constants α, β, γ , δ satisfy a single relation β 2 − 4αγ + 12δ = 0. These

densities were first introduced in [27], where it was shown that the corresponding
Hamiltonian chains possess an infinity of conservation laws which Poisson commute
and form a complete system. One can prove that all other components H ijk , i ≥ 2, of
the Haantjes tensor are identically zero.
The aim of this section is to characterize all densities of the form h(u 1 , u 2 , u 3 )
such that the Haantjes tensor of the Hamiltonian chain (13) is zero. As in the previous
Sections, the conditions H jk 1 = 0 provide expressions for all of the ten third order
partial derivatives of h, the simplest six of them being
5h 233 5h 13 h 33 5h 23 h 33
h 333 = , h 133 = , h 233 = ,
2h 3 2h 3 2h 3
3h 213 + 2h 33 h 11
h 113 = , (29)
2h 3
3h 13 h 23 + 2h 33 h 12
h 123 = ,
2h 3
3h 223 + 2h 33 h 22
h 223 = ;
2h 3
the remaining expressions for h 111 , h 112 , h 122 , h 222 are not written out explicitly due
to their complexity (we have used Mathematica to manipulate with these expres-
sions). It was verified that the system for h is in involution, and its solution space
is 10-dimensional. In what follows, (29) refers to the full system of ten equations. It
was demonstrated recently in [21] that the relations (29) imply the vanishing of all
remaining components of the Haantjes tensor, that is,
1
H jk = 0 ⇒ H ijk = 0 for any i ≥ 2.
Before proceeding with a detailed analysis of the equations (29), let us make a digres-
sion on conservation laws of the chains (13). First of all, any Hamiltonian chain (13)
possesses three conservation laws of the form
u 1t = (u 1 h 2 + 2u 2 h 3 )x ,
u 2t = (u 1 h 1 + 2u 2 h 2 + 3u 3 h 3 − h)x ,
and
h t = (u 1 h 1 h 2 + 2u 2 h 1 h 3 + u 2 h 22 + 3u 3 h 2 h 3 + 2u 4 h 23 )x ,
123
which correspond to the conservation of the Casimir, momentum and the Hamiltonian,
respectively. Let us require the existence of an extra ‘higher’ conservation law of the
form
P(u 1 , u 2 , u 3 , u 4 )t = Q(u 1 , u 2 , u 3 , u 4 , u 5 )x (30)
whose density P depends on the first four coordinates u i (the structure of the flux Q
follows from the equations of the chain and does not constitute an additional restric-
tion). We have obtained the following
Proposition The relations (29) are necessary and sufficient for the existence of an
extra conservation law of the form (30).
The proof is computational, see Appendix 2 for details. In fact, we have the follow-
ing stronger result
Theorem 4 The relations (29) imply the existence of a generating function which
gives rise to an infinity of conservation laws. To be more precise, we claim that for
any n there exist n linearly independent conserved densities which are functions of the
first n coordinates u i . All higher conservation laws are polynomial in u 4 , u 5 , u 6 , . . ..
The proof of this statement is given in Sect. 4.2 below.
4.1 Integration of the system (29)
5/2
The first three equations in (29) imply that h 33 = sh 3 , s = const. Thus, there are
two cases to consider, s = 0 and s = 0.
Case I: s = 0
In this case h is linear in u 3 and the Eqs. (29) imply
u3
h= + p(u 1 , u 2 ).
(c + au 1 + bu 2 )2
This ansatz can be simplified by utilizing the canonical transformations [29],
1 3 1
ũ 1 = λu 1 , ũ 2 = u 2 , ũ 3 = u , ũ 4 = 2 u 4 , . . . , (31)
λ λ
etc, λ=const, and
ũ 1 = u 1 , ũ 2 = u 2 + su 1 , ũ 3 = u 3 + 2su 2 + s 2 u 1 , . . . , (32)
etc, s=const. Both transformations preserve the Poisson bracket specified by (13).
Hence, they can be used to simplify the Hamiltonian. Suppose, for instance, that
b = 0. Then, up to the second canonical transformation, one can assume the ansatz
h = u 3 /(c + u 2 )2 + p(u 1 , u 2 ). If b = 0 then h = u 3 /(c + u 1 )2 + p(u 1 , u 2 ). Thus,
there are two subcases:
123
Subcase I1 : h = u 3 /(c + u 1 )2 + p(u 1 , u 2 ). The substitution of this ansatz into the

four remaining equations for h implies the following system for p:
16(u 2 )2 +(c+u 1 )3 (8(u 2 )2 p22 +4(3c+u 1 )u 1 p12 )+2(c+u 1 )4 (5c+u 1 ) p11

p111 = −
c(c+u 1 )5 (2+u 1 (c+u 1 )2 p22 )
(c + u )(c p12 − c(c − 3u 1 ) p22 p11 )
1 2 2
+ ,
c(2 + u 1 (c + u 1 )2 p22 )
2u 2 − c(c + u 1 )3 p12 2 + c(c + u 1 )2 p22
p112 =2 , p 122 = − , p222 = 0.
c(c + u 1 )4 c(c + u 1 )3
) 2 21 βu 2
The last three equations lead to p(u 1 , u 2 ) = 1+αc(c+u
c(c+u 1 )2
(u ) + c+u 1 + q(u ) where
1
α and β are arbitrary constants. The substitution into the first equation gives a linear
ODE for q,
c2 β 2
(αc2 − 3u 1 αc − 4)q − (c + u 1 )(1 + cu 1 α)q = .
2(c + u 1 )3
Without any loss of generality one has

m n
q(u 1 ) = +
(c + u 1 )2 c + u1
where the constants m, n satisfy a single relation 2(1 − c2 α)n − 6cαm + 21 c2 β 2 = 0.

Ultimately, we have Hamiltonian densities
1 + αc(c + u 1 ) 2 2 βu 2 m n
h = u 3 /(c + u 1 )2 + (u ) + + + .
c(c + u 1 )2 c + u1 (c + u 1 )2 c + u1
Notice that the constant β can be eliminated by the second canonical transformation.
Subcase I2 : h = u 3 /(c + u 2 )2 + p(u 1 , u 2 ). The substitution of this ansatz into the
four remaining equations for h implies the following system for p,
(u 1 )3 (c+u 2 ) p111 −2c(u 1 )2 p11 −2cu 1 (c−u 2 ) p12 −(5c−3u 2 )(c−u 2 )2 p22
p222 = ,
(c − u 2 )3 (c + u 2 )
(u 1 )2 (c + u 2 ) p111 − 2cu 1 p11 − 2(2c2 − 3cu 2 + (u 2 )2 ) p12
p122 = ,
(c − u 2 )2 (c + u 2 )
u 1 (c + u 2 ) p111 + (u 2 − 3c) p11
p112 = ,
(c − u 2 )(c + u 2 )
along with one more relation for p111 which we do not write out due to its complexity.
u 1 u 2 −αu 2 − αc
3 +(c−u ) g(η)
2 3
The first three relations imply p(u 1 , u 2 ) = (c+u 2 )2
where g(η) is an
1
arbitrary function of a single variable η = u
c−u 2
. The result of the substitution of this
ansatz into the remaining equation for p111 factors into a product of two terms, leading
to the cases (i) and (ii) below:
123
(i) The function g satisfies a first order ODE
4c2 η2 g − 12c2 ηg − 1 − αη + 2cη2 = 0,
η α
the solution of this equation is g(η) = µη3 + 4c − 12c2
− 1
16c2 η
. This results in the
Hamiltonian densities of the form
u3 u1u2 u 2 + c/3 (u 1 )3
h= + −α +µ
(c + u )
2 2 (c + u )
2 2 (c + u )
2 2 (c + u 2 )2

1 (c − u 2 )3 u1 α c − u2
+ − − .
4c (c + u 2 )2 c − u 2 3c 4cu 1
(ii) The function g solves a third order ODE
(4c2 η2 g − 12c2 ηg − 1 − αη + 2cη2 )g + (α − 4cη + 12c2 g + 4c2 ηg

1
−2c2 η2 g )g + (4c − 8c2 g )g − = 0.
2
Remarkably, this complicated equation trivializes after being differentiated by η once,

taking the form
(4c2 η2 g − 12c2 ηg − 1 − αη + 2cη2 )g = 0.
Since the possibility when the coefficient at g equals zero was considered above in
the case (i), we conclude that g must be a cubic polynomial,
g = µ + νη + γ η2 + δη3
where the constants satisfy a single relation 12δ−8cν+16c2 (ν 2 −3γ µ)−4γ α+1 = 0
which can be obtained by substituting back into the original ODE. This leads to
Hamiltonian densities of the form
u3
h=
(c + u 2 )2
αc
u 1 u 2 − αu 2 − + µ(c − u 2 )3 + νu 1 (c − u 2 )2 + γ (u 1 )2 (c − u 2 ) + δ(u 1 )3
+ 3
.
(c + u 2 )2
Notice that the particular case c = 0 results (up to obvious equivalence transformations
and relabeling of constants) in Hamiltonian densities of the form
u3 (u 1 )2 u1 1 (u 1 )3
h= + α + β + γ + δ (33)
(u 2 )2 u2 u2 u2 (u 2 )2
where α, β, γ , δ satisfy a single relation β 2 − 4αγ + 12δ = 0.
123
Remark The apparent similarity of the cases (28) and (33) is not accidental and mani-
fests an important reciprocal invariance of the class of Hamiltonian chains (13). Recall
that the conservation of momentum reads
u 2t = (u 1 h 1 + 2u 2 h 2 + 3u 3 h 3 − h)x .
Let us change from x, t to the new independent variables X, T where
d X = u 2 d x + (u 1 h 1 + 2u 2 h 2 + 3u 3 h 3 − h)dt, T = t.
It is known that reciprocal transformations of this type preserve Poisson brackets of

the form (13), see e.g. [9] (it is crucial that u 2 is the momentum of the correspond-
ing Poisson bracket). One can verify directly that performing the above change of
independent variables and introducing
h u1 1 u3 un
H= , U 1
= , U 2
= , U 3
= , . . . , U n
= , (34)
u2 u2 u2 (u 2 )3 (u 2 )n
one arrives at the system which takes the original form (13) in the variables X, T,
U n , H . Thus, the above reciprocal transformation is canonical. One can verify directly
that the change of variables (34) indeed identifies (28) and (33).
Case II: s = 0
In this case the elementary integration gives
h = (γ u 3 + p(u 1 , u 2 ))1/3 + q(u 1 , u 2 ),
and the substitution into the last three equations (29) implies that q is linear. Up to the
equivalence h → αh + au 1 + bu 2 + c we thus have
h = (u 3 + p(u 1 , u 2 ))1/3 .
The substitution of this ansatz into the remaining equations for h implies a complicated
system of third order PDEs for p(u 1 , u 2 ). A useful observation is that this system is
invariant under a 3-parameter group of point symmetries which is generated by the
two canonical transformations (31, 32) and the reciprocal transformation (34). The
infinitesimal generators of these symmetries are
X 1 = u 1 ∂u 1 − p∂ p , X 2 = u 1 ∂u 2 − 2u 2 ∂ p ,
X 3 = (u 1 )2 ∂u 1 + u 1 u 2 ∂u 2 + (3 pu 1 + 2(u 2 )2 )∂ p ;
they satisfy the commutator relations
[X 1 , X 2 ] = X 2 , [X 1 , X 3 ] = X 3 , [X 2 , X 3 ] = 0.
123
These symmetries suggest a change of variables which considerably simplifies the

equations for p. The idea is to choose new coordinates such that the symmetry genera-
tors assume the simplest possible form. Introducing ξ = − u11 , η = uu 1 , s = pu (u+(u
2 1 2 )2
1 )4 ,
we have
X 1 = −ξ ∂ξ − η∂η − 4s∂s , X 2 = ∂η , X 3 = ∂ξ .
Setting s = s(ξ, η) we have p = s(ξ, η)(u 1 )3 − (uu 1) . In terms of s(ξ, η) the equations
2 2
for p assume a remarkable symmetric form
sξ ξ η sηη − sηηη sξ ξ = 4sη , (35)

sξ ηη sξ ξ − sξ ξ ξ sηη = 4sξ , (36)
sηηη sξ + sξ ηη sη = sηη sξ η , (37)
sξ ξ ξ sη + sξ ξ η sξ = sξ ξ sξ η , (38)
1
sηηη sξ ξ sη + sξ ηη sηη sξ + 4sη2 = sηη (12s + sξ2η + sξ ξ sηη ), (39)
2
1
sξ ξ ξ sηη sξ + sξ ξ η sξ ξ sη + 4sξ = sξ ξ (12s + sξ2η + sξ ξ sηη );
2
(40)
2
in the process of derivation of these equations we have assumed that sξ , sη and

sξ ξ sη2 − sηη sξ2 are nonzero: these expressions appear as denominators in the interme-
diate calculations. Particular cases when some of these expressions vanish will be
discussed below.
To solve the Eqs. (35–40) we proceed as follows. Differentiating (37), (38) and
using (35), (36) we obtain four relations among the fourth order partial derivatives
of s,
sξ sηηηη + sη sξ ηηη = 0, sξ sξ ηηη + sη (sξ ξ ηη − 4) = 0,

sξ (sξ ξ ηη − 4) + sη sξ ξ ξ η = 0, sξ sξ ξ ξ η + sη sξ ξ ξ ξ = 0,
which can be parametrized as
sηηηη = q, sξ ηηη = qr, sξ ξ ηη − 4 = qr 2 , sξ ξ ξ η = qr 3 , sξ ξ ξ ξ = qr 4 , (41)
r = −sξ /sη . The further analysis leads to the two possibilities.

Subcase II1 : q = 0. In this case s(ξ, η) is a polynomial of the form
s = ξ 2 η2 + aξ 3 + bξ 2 η + cξ η2 + dη3 + αξ 2 + βξ η + γ η2 + µξ + νη + .
The substitution into the remaining equations for s implies the following relations
among the coefficients:
β = bc − 9ad, µ = αc − 3aγ , ν = bγ − 3αd, 12 + β 2 + 4αγ = 4(bν + cµ).
123
Notice that the corresponding p(u 1 , u 2 ) = s(− u11 , uu 1 )(u 1 )3 − (uu 1) will be a cubic
2 2 2
polynomial in u 1 , u 2 . A particular example from this class with the Hamiltonian den-
sity h = (u 3 + τ )1/3 was discussed in [27]. It corresponds to the case where a = −τ
and all other coefficients of s(ξ, η) are zero.
Subcase II2 : q = 0. Then the consistency conditions of (41) imply the relations qξ =
(qr )η , rξ = rrη . Taking into account that r = −sξ /sη we have sξ ξ sη2 − sηη sξ2 = 0.
This case is discussed below.
There are two more possibilities one needs to consider to complete the classification
(notice that the Eqs. (35–40) can no longer be used since they were derived under the
assumption that certain expressions do not vanish).
Subcase II3 : sξ = 0 or sη = 0. A simple analysis leads to Hamiltonian densities of
the form
1/3
(u 2 )2
h = u + α(u ) − 1
3 1 3
,
u
α =const.
s
Subcase II4 : sξ ξ sη2 − sηη sξ2 = 0. Setting r = − sηξ one obtains
rξ = rrη , sξ = −r sη .
Calculating partial derivatives of s with the help of the above relations and substituting
1 = 0 we obtain r
them into the conditions H jk ηη = 0; without any loss of generality
η
one can set r = − ξ . This implies ξ sξ − ηsη = 0, therefore, s = s(z), z = ξ η =
−u 2 /(u 1 )2 . For s(z) we obtain the ODE
8z 2 s s + 8zs s − 4z 2 (s )2 − (s )2 − 12s = 0
which linearizes after being differentiated by z once. Its general solution is given by
the formula
s(z) = z 2 + α + β(−z) + γ (−z)1/2 + δ(−z)3/2
where the constants α, β, γ , δ satisfy a single quadratic relation 12α + β 2 − 3γ δ = 0.

This results in Hamiltonian densities of the form
1/3
h = u 3 + α(u 1 )3 + βu 1 u 2 + γ (u 1 )2 (u 2 )1/2 + δ(u 2 )3/2 .
4.2 Generating functions for conservation laws
In this section we demonstrate that any Hamiltonian chain (13) whose density h sat-
isfies the system (29), possesses a generating function which provides an infinity of
conservation laws. We refer to [15] for a calculation of generating functions for a class
of integrable Hamiltonian chains associated with Kupershmidt’s brackets. To illustrate
the method of generating functions let us consider the Benney chain (2) and introduce
the generating function
123
u1 u2 u3
λ= p+ + 2 + 3 + ··· (42)
p p p
which was shown in [17] to satisfy, by virtue of (2), the fundamental relation

λt − pλx = λ p pt − ( p 2 /2 + u 1 )x . (43)
This relation provides an infinity of conserved densities Hn (u) defined by the equation
pt = ( p 2 /2 + u 1 )x
where one has to substitute the expression for p(λ) obtained from (42): p = λ − Hλ1 −
H2
λ2
−H 3
λ3
− · · · . Explicitly, one gets H1 = u 1 , H2 = u 2 , H3 = u 3 + (u 1 )2 , etc.
Given an arbitrary Hamiltonian density h(u 1 , u 2 , u 3 ), a generating function is
sought in the form
∞ k
u
λ = ψ( p, u 1 , u 2 ) + ; (44)
pk
k=1
notice that the dependence on higher moments u 3 , u 4 , u 5 , etc, is exactly the same
as in (42). As demonstrated in [41], the generating function (44) has to satisfy the
Gibbons-type relation

λt − (2 ph 3 + h 2 )λx = λ p pt − ( p 2 h 3 + ph 2 + h 1 )x , (45)
which reduces to (42) for the Benney Hamiltonian h = (u 3 + (u 1 )2 )/2. Substituting

(44) into (45) one obtains, by virtue of (13), the following relations among the first
order partial derivatives ψ1 , ψ2 , ψ p :
3u 3
(ψ p − )( p 2 h 13 + ph 12 + h 11 ) + ψ1 (2u 2 h 13 + u 1 h 12 − 2 ph 3 ) + ψ2 (3u 3 h 13
p4
1 3u 3
+2u 2 h 12 + u 1 h 11 ) + 3 (3u 3 h 12 + (2u 2 + )h 11 ) = 0,
p p
3u 3
(ψ p − )( p 2 h 23 + ph 22 + h 12 ) + ψ1 (2u 2 h 23 + u 1 h 22 + 2h 3 ) (46)
p4
1 3u 3
+ψ2 (3u 3 h 23 + 2u 2 h 22 + u 1 h 12 − 2 ph 3 ) + 3 (3u 3 h 22 + (2u 2 + )h 12 ) = 0,
p p
3u 3
(ψ p − 4 )( p 2 h 33 + ph 23 + h 13 ) + ψ1 (2u 2 h 33 + u 1 h 23 ) + ψ2 (3u 3 h 33 + 2u 2 h 23
p
1 3u 3
+u 1 h 13 + 2h 3 ) + 3 (3u 3 h 23 + (2u 2 + )h 13 − 2 ph 3 ) = 0.
p p
123
These relations uniquely specify the partial derivatives ψ1 , ψ2 , ψ p :
1 2
ψ1 = − 4 p h 3 ( ph 3 h 33 + h 23 (2h 3 + u 1 h 13 ) − h 33 (u 1 h 12 + u 2 h 22 ))
× p(2h 3 (3u 3 (h 33 h 22 − h 223 ) + 2h 3 (h 22 + h 13 ) + u 1 (h 13 (2h 22 + h 13 )

−2h 12 h 23 − h 11 h 33 )) + (u 1 )2 (h 12 (h 12 + h 33 − h 13 h 23 ) + h 23 (h 11 h 23
−h 12 h 13 ) + h 22 (h 213 − h 11 h 33 )))h 3 (6u 3 (h 13 h 23 − h 12 h 33 ) − 2h 3 h 12
+u 1 (h 11 h 23 − h 12 h 13 )) + 2u 2 u 1 (h 12 (h 12 h 33 − h 13 h 23 ) + h 23 (h 11 h 23
−h 12 h 13 ) + h 22 (h 212 − h 11 h 33 )) ,
1
ψ2 = − 4 ph 23 h 33 p 2 + 4h 3 (h 23 (h 3 + u 2 h 23 + u 1 h 13 ) − h 33 (u 2 h 22 + u 1 h 12 )) p
+(u 1 )2 (h 12 (h 12 h 33 − h 13 h 23 ) + h 23 (h 11 h 23 − h 12 h 13 ) + h 22 (h 213 − h 11 h 33 ))
+2h 3 (h 13 (2h 3 + 2u 2 h 23 + u 1 (h 13 + h 22 )) − h 12 (2u 2 h 33 + u 1 h 23 )
−u 1 h 11 h 33 ) ,
1 2
ψp = − 4h 3 h 33 u 1 p 2 + 4h 3 p(2h 3 (u 2 h 33 + u 1 h 23 ) + u 2 u 1 (h 223 − h 22 h 33 )
−(u 1 )2 (h 13 h 23 − h 12 h 33 )) + 2h 3 (2h 3 (3u 3 h 33 + 4u 2 h 23 + u 1 (2h 13 + h 22 )

+(u 1 )3 (h 12 (h 12 h 33 − h 13 h 23 ) + h 23 (h 23 h 11 − h 12 h 13 ) + h 22 (h 213 − h 11 h 33 )))
+4(u 2 )2 (h 223 − h 22 h 33 ) + 4u 1 u 2 (h 13 h 23 − h 12 h 33 ) − 3u 1 u 3 (h 223 − h 22 h 33 )
+(u 1 )2 (h 13 (2h 22 + h 13 ) − 2h 23 h 12 − h 11 h 33 )) .
= 4 p 4 h 23 h 33 + 4 p 3 h 2 (2h 3 h 23 + u 2 (h 223 − h 22 h 33 ) + u 1 (h 13 h 23 − h 12 h 33 ))
+((u 1 )2 p 2 + 2u 2 u 1 p)(2h 3 (h 13 (h 13 + 2h 22 ) − 2h 12 h 23 − h 11 h 33 ) + (h 12
×(h 12 h 33 − h 13 h 23 ) + h 23 (h 11 h 23 − h 12 h 13 )h 22 (h 213 − h 11 h 33 ))) + 2h 3 p 2
×(3u 3 (h 22 h 33 − h 223 ) + 2h 3 (2h 13 + h 22 )) + 4h 3 p(3u 3 (h 13 h 23 − h 12 h 33 )
−2h 3 h 12 ) + 2h 11 (h 3 (2h 3 + u 1 h 22 + 4u 2 h 23 ) + 2(u 2 )2 (h 223 − h 22 h 33 ))
−3u 3 ((h 213 − h 11 h 33 )(2h 3 + u 1 h 22 ) + u 1 h 23 (h 12 h 33 − h 13 h 23 )
+u 1 h 23 (h 11 h 23 − h 12 h 13 )).
These equations are consistent provided the density h satisfies the system (29). This
proves Theorem 4: indeed, an infinite sequence of conservation laws results from the
equation
pt = ( p 2 h 3 + ph 2 + h 1 )x
where one has to substitute the expansion for p in terms of λ obtained from (44).
The calculation of ψ( p, u 1 , u 2 ) can be summarized as follows:
123
– one first integrates the equation for ψ p , which appears to be rational in p, with
respect to p. This defines ψ up to a function of u 1 , u 2 .
– one fixes this function by a substitution into the equations for ψ1 , ψ2 .
This procedure leads to the following explicit formulae for generating functions for
the Hamiltonian densities obtained in Sect. 4.1.
Linear case. This is the simplest generalization of the Benney Hamiltonian with the
density
h(u 1 , u 2 , u 3 ) = u 3 + α(u 1 )2 + βu 1 u 2 + γ (u 2 )2 + δ(u 1 )3 ,
where the constants satisfy the condition β 2 − 4αγ + 12δ = 0. Even in this case the
function ψ is quite nontrivial,
2 β + 2γ p
ψ( p, u 1 , u 2 ) = 1
arctan 1
.
(4αγ − β2) 2 (1 + γ u 1 )(4αγ − β 2 ) 2
The generating function is obtained by substituting this expression into (44).

Subcase I1 . In this case the Hamiltonian density takes the form
u3 1 + αc(c + u 1 ) 2 2 βu 2 m n
h(u 1 , u 2 , u 3 ) = + (u ) + + + ,
(c + u )
1 2 c(c + u )
1 2 c + u1 (c + u 1 )2 c + u 1
where the constants satisfy the single relation, 2(1 − c2 α)n − 6cαm + 21 c2 β 2 = 0.
The corresponding function ψ is given by
3
c2 2(c2 α − 1)u 2 + c(2(c2 α − 1) p − βc) − cβu 1
ψ( p, u 1 , u 2 ) = − 1
tanh−1 1
.
(3m) 2 2(3cm) 2 (1 + cαu 1 )
Subcase I2 (i). For the Hamiltonian density
u 3 + u 1 u 2 − α(u 2 + 3c ) + µ(u 1 )3 (c − u 2 )3
h(u 1 , u 2 , u 3 ) = +
(c + u 2 )2 4c(c + u 2 )2

u 1 α c−u 2
× − − ,
c−u 2 3c 4cu 1
the function ψ is given by
1 (u 2 )2 − 2cu 2 + 4cpu 1 + c2
ψ( p, u 1 , u 2 ) = 1
arctan 1
.
(3µ) 2 4c(3µ) 2 (u 1 )2
Subcase II3 : For the Hamiltonian density
(u 2 )2 1
h(u 1 , u 2 , u 3 ) = (u 3 + α(u 1 )3 − )3 ,
u1
123
the function ψ is given by
1 pu 1 − u 2
ψ( p, u 1 , u 2 ) = 1
arctan 1
.
(3α) 2 (3α) 2 (u 1 )2
Other examples obtained in Sect. 4.1 lead to more complicated expressions for ψ.
5 Hydrodynamic reductions and the diagonalizability
To illustrate the method of hydrodynamic reductions we consider the Benney chain (2),
u 1t = u 2x ,
u 2t = u 3x + u 1 u 1x ,
u 3t = u 4x + 2u 2 u 1x ,
u 4t = u 5x + 3u 3 u 1x ,
etc. Following the approach of [19,20] let us seek solutions in the form u i = u i
(R 1 , . . . , R m ) where the Riemann invariants R 1 , . . . , R m solve a diagonal system
Rti = λi (R)Rxi .
Substituting this ansatz into the Benney equations and equating to zero coefficients at
Rxi we arrive at the following relations:
∂i u 2 = λi ∂i u, (47)
∂i u 3 = ((λi )2 − u)∂i u, (48)
∂i u 4 = ((λi )3 − uλi − 2u 2 )∂i u, (49)
∂i u 5 = ((λi )4 − u(λi )2 − 2u 2 λi − 3u 3 )∂i u, (50)
etc. Here u = u 1 , ∂i = ∂ R i , i = 1, . . . , m (no summation!) The consistency condi-

tions of the first three relations (47–49) imply
∂ j λi ∂i λ j
∂i ∂ j u = ∂i u + ∂ j u,
λ j − λi λi − λ j
∂ j λi ∂i u + ∂i λ j ∂ j u = 0,
λi ∂ j λi ∂i u + λ j ∂i λ j ∂ j u + ∂i u∂ j u = 0,
respectively. Solving these equations for ∂ j λi we arrive at the Gibbons–Tsarev system
∂ju ∂i u∂ j u
∂ j λi = , ∂i ∂ j u = 2 i . (51)
λ j − λi (λ − λ j )2
123
It is a truly remarkable fact that all other consistency conditions (e.g., of the relation
(50), etc), are satisfied identically modulo (51). Moreover, the semi-Hamiltonian prop-
erty (9) is also automatically satisfied. Thus, the system (51) governs m-component
reductions of the Benney chain. Up to reparametrizations R i → f i (R i ) these reduc-
tions depend on m arbitrary functions of a single variable. Solutions arising within
this approach are known as multiple waves, or nonlinear interactions of planar simple
waves.
The above approach clearly applies to any hydrodynamic chain from the class C.
Let us restrict, for instance, to the chains of the type (6). Looking for solutions in the
form u i = u i (R 1 , . . . , R m ) and substituting this ansatz into (6) we arrive at an infinite
system of relations similar to (47–50). The first three of them imply the ‘generalized
Gibbons–Tsarev system’ of the form
∂ j λi = (. . .)∂ j u, ∂i ∂ j u = (. . .)∂i u∂ j u,
u = u 1 , where dots denote complicated expression which are rational in λi with the
coefficients depending on the chain under study (that is, on g, h, etc). Requiring that all
other consistency conditions, as well as the semi-Hamiltonian property, are satisfied
identically we obtain constraints for the matrix V (u), see [10–14,16] where a similar
approach was applied to the classification of integrable multi-dimensional quasilin-
ear systems. The main result of this section is the proof of the following theorem
formulated in the introduction:
Theorem 2 The vanishing of the Haantjes tensor is a necessary condition for the
existence of infinitely many hydrodynamic reductions and, thus, for the integrability
of a hydrodynamic chain.
We will give two different proofs of this statement. Based on essentially different
ideas, they may be of interest in their own right.
First Proof This proof is computational. Writing down the equations of the chain in
the form u mt = Vn u x and substituting the ansatz u = u (R , . . . , R ) we arrive at
m n i i 1 m
an infinite set of relations
Vnm ∂i u n = λi ∂i u m ;
we point out that all summations here and below involve finitely many nonzero terms.
Applying the operator ∂ j , j = i, we obtain
m
Vn,k ∂i u n ∂ j u k + Vnm ∂i ∂ j u n = ∂ j λi ∂i u m + λi ∂i ∂ j u m . (52)
Interchanging the indices i and j and subtracting the results we arrive at the expression
for ∂i ∂ j u m in the form
m − Vm
∂ j λi ∂i λ j Vn,k k,n
∂i ∂ j u m = ∂i u m
+ ∂ j u m
+ ∂i u n ∂ j u k .
λ −λ
j i λ −λ
i j λ −λ
i j
123
Substituting this back into (52) we arrive at a simple relation

m ∂ un ∂ uk
Nnk i j
∂ j λi ∂i u m + ∂i λ j ∂ j u m =
λ −λ
i j
where N is the Nijenhuis tensor of V . This can be rewritten in the invariant form
N (∂i u, ∂ j u)
∂ j λi ∂i u + ∂i λ j ∂ j u = (53)
λi − λ j
which implies the following four relations:
V 2 N (∂i u, ∂ j u)
(λi )2 ∂ j λi ∂i u + (λ j )2 ∂i λ j ∂ j u = λi −λ j
,
V N (V ∂i u, ∂ j u)
(λi )2 ∂ j λi ∂i u + λi λ j ∂i λ j ∂ j u = λi −λ j
,
V N (∂i u, V ∂ j u)
λi λ j ∂ j λi ∂i u + (λ j )2 ∂i λ j ∂ j u = λi −λ j
,
N (V ∂i u, V ∂ j u)
λi λ j ∂ j λi ∂i u + λi λ j ∂i λ j ∂ j u = λi −λ j
.
For instance, the first relation can be obtained by applying the operator V 2 to (53) and
using V ∂i u = λi ∂i u. Thus,
V 2 N (∂i u, ∂ j u) − V N (V ∂i u, ∂ j u) − V N (∂i u, V ∂ j u) + N (V ∂i u, V ∂ j u) = 0.
The last relation can be rewritten in the form H (∂i u, ∂ j u) = 0 where H is the Haantjes
tensor, indeed, a coordinate-free form of the relation (11) is
H (X, Y ) = V 2 N (X, Y ) − V N (V X, Y ) − V N (X, V Y ) + N (V X, V Y )
where X, Y are arbitrary vector fields. Keeping in mind that (a) ∂i u and ∂ j u are
eigenvectors of the matrix V corresponding to the eigenvalues λi and λ j , and (b) λi
and λ j can take arbitrary values, we conclude that H (X, Y ) = 0 for any two formal
eigenvectors of the matrix V .
Assuming that formal eigenvectors of V span the space of dependent variables u
(this is true for all examples discussed in this paper), we obtain H = 0. In more
detail, let X (λ) = (ξ 1 (λ), ξ 2 (λ), ξ 3 (λ), . . .)t be a formal eigenvector of the matrix
V corresponding to the eigenvalue λ. Let us assume that these eigenvectors span the
space of dependent variables, that is, that there exist no non-trivial relations of the
form ci ξ i (λ) = 0 with finitely many nonzero λ-independent coefficients ci . In other
words, ξ i (λ) are linearly independent as polynomials in λ.
The condition H (X (λ), X (µ)) = 0, written in components, takes the form
H ijk ξ j (λ)ξ k (µ) = 0; recall that, since the matrix V belongs to the class C, these
sums contain finitely many terms for any fixed value of the upper index i. Taking into
account the linear independence of ξ j (λ) and ξ k (µ), one readily arrives at H ijk = 0.
123
As an illustration, let V be the matrix corresponding to the Benney chain (2). Then
X (λ) = (1, λ, λ2 − u 1 , λ3 − u 1 λ − 2u 2 , . . .)t , so that each ξ i (λ) is a polynomial in
λ of the degree i − 1. These eigenvectors span the space of dependent variables since
polynomials ξ i (λ) are manifestly linearly independent.
Notice that we have proved a more general result, namely, that the existence of
sufficiently many two-component reductions already implies the vanishing of the
Haantjes tensor H . Indeed, nothing changes in the proof if we set i = 1, j = 2 in the
formula (53). As demonstrated in the Theorem 3 below, one can further strengthen
the result by reversing the above proof under the additional assumption of the
simplicity of the spectrum of the matrix V . This will require the relations
N (V ∂i u, ∂ j u) − V N (∂i u, ∂ j u)
∂i λ j ∂ j u = ,
(λi − λ j )2
N (V ∂ j u, ∂i u) − V N (∂ j u, ∂i u)
∂ j λi ∂i u = ,
(λi − λ j )2
which can be obtained by applying V to both sides of (53) and solving for ∂i λ j ∂ j u
and ∂ j λi ∂i u.
Second Proof Our first remark is that for the chains from the class C one needs to
know only finitely many rows of the matrix V (u) to calculate each particular com-
ponent of the Haantjes tensor. Let us fix the values of indices i, j, k and denote by
C(i, j, k) the maximal number of rows needed to calculate H ijk (counting from the
first row). We need to show that H ijk = 0. Let us consider an m-component diagonal
reduction u i (R 1 , . . . , R m ), i = 1, 2, . . .. Choosing the first m variables u 1 , . . . , u m
as independent, we can represent the reduction explicitly as
u m+1 = u m+1 (u 1 , . . . , u m ), u m+2 = u m+2 (u 1 , . . . , u m ), . . . ,
etc. Substituting these expressions into the first m equations of the chain we obtain
an m-component system Sm for u 1 , . . . , u m , while the remaining equations will be
satisfied identically (by the definition of a reduction). Notice that the Haantjes tensor
of the reduced system Sm is identically zero since the reduction is diagonalizable. Let
us now choose the number m sufficiently large so that the first C(i, j, k) equations of
the chain do not contain variables u m+1 , u m+2 , . . . (one can always do so since any
equation of the chain depends on finitely many u’s, and m can be arbitrarily large).
Then the first C(i, j, k) equations of the reduced system Sm will be identical to the
first C(i, j, k) equations of the original infinite chain. Hence, the corresponding com-
ponents H ijk for the reduced system and for the infinite chain will also coincide. This
proves that all components of the Haantjes tensor of the chain must be zero.
A straightforward modification of the second proof allows one to show that the exis-
tence of an infinity of semi-Hamiltonian reductions implies the vanishing of the tensor
P. This establishes the necessity of the conjecture formulated in the Introduction.
We emphasize that the condition of diagonalizability alone is not sufficient for the
integrability in general. This can be seen as follows.
123
Example Let us consider the chain
u 1t = u 2x + p(u 1 )u 1x ,
u 2t = u 3x + p(u 1 )u 2x + u 1 u 1x ,
u 3t = u 4x + p(u 1 )u 3x + 2u 2 u 1x ,
u 4t = u 5x + p(u 1 )u 4x + 3u 3 u 1x ,
etc, which is obtained from the Benney chain ut = V (u)ux by the transformation
V → V + p(u 1 )E where E is an infinite identity matrix and p is a function of u 1 .
One can verify that the corresponding Haantjes tensor is zero (which is not at all
surprising since the addition of a multiple of the identity does not effect the diagonal-
izability). A simple calculation shows that hydrodynamic reductions of this chain are
governed by exactly the same equations as in the Benney case, the only difference is
that now the Riemann invariants R i solve the equations
Rti = (λi (R) + p(u 1 ))Rxi .
The semi-Hamiltonian property is satisfied if and only if p = 0. Thus, we have con-

structed examples which possess infinitely many diagonal hydrodynamic reductions
none of which are semi-Hamiltonian (if p = 0).
Theorem 2 can be strengthened under the additional assumption of the simplicity

of the spectrum of V .
Theorem 3 In the simple spectrum case the vanishing of the Haantjes tensor H is
necessary and sufficient for the existence of two-component reductions parametrized
by two arbitrary functions of a single argument.
Proof The necessity part is contained in the first proof of Theorem 2. To establish
the sufficiency one has to show that the vanishing of the Haantjes tensor implies the
solvability of the equations
V ∂1 u = λ1 ∂1 u, V ∂2 u = λ2 ∂2 u, (54)
m − Vm
∂2 λ1 ∂1 λ 2 Vn,k k,n
∂1 ∂2 u m = ∂ 1 u m
+ ∂ 2 u m
+ ∂1 u n ∂2 u k (55)
λ2 − λ1 λ1 − λ2 λ1 − λ2
and
N (V ∂1 u, ∂2 u)−V N (∂1 u, ∂2 u)
∂1 λ2 ∂2 u = (λ1 −λ2 )2
,
(56)
N (V ∂2 u, ∂1 u)−V N (∂2 u, ∂1 u)
∂2 λ1 ∂ 1u = (λ1 −λ2 )2
,
which govern two-component reductions. Our first observation is that the vanishing
of the Haantjes tensor implies that the vectors N (V ∂1 u, ∂2 u) − V N (∂1 u, ∂2 u) and
N (V ∂2 u, ∂1 u) − V N (∂2 u, ∂1 u) are the eigenvectors of V with the eigenvalues λ2
and λ1 , respectively. By the assumption of the simplicity of the spectrum, they are
123
proportional to ∂2 u and ∂1 u. Thus, equations (56) reduce to a pair of first order PDEs
for λ1 and λ2 ,
k1 k2
∂1 λ2 = 1 , ∂2 λ1 = 1 , (57)
(λ − λ )2 2 (λ − λ2 )2
here k1 and k2 are the corresponding coefficients of proportionality. The relations (54)
allow one to reconstruct all components u 2 , u 3 , . . . of the infinite vector u in terms of
its first component u 1 . Finally, the Eqs. (55), which are nothing but the consistency
conditions of (54), reduce to a single second order PDE for the first component u 1 of the
infinite vector u. Thus, relations (54–56) reduce to a pair of first order Eqs. (57) plus one
second order PDE for u 1 . Up to reparametrizations R 1 → f 1 (R 1 ), R 2 → f 2 (R 2 ),
their general solution depends on two arbitrary functions of a single variable.
6 Conclusion
We have proposed a simple and easy-to-verify necessary condition for the integrability
of hydrodynamic chains based on the vanishing of the Haantjes tensor. Conservative
and Hamiltonian examples are discussed, illustrating the general approach. We con-
jecture that all examples arising in Sects. 3 and 4 from the requirement of the vanishing
of the Haantjes tensor (and, thus, satisfying a necessary condition for the integrability)
also satisfy the following properties:
(i) they possess infinitely many m-component hydrodynamic reductions parametrized
by m arbitrary functions of a single variable (thus, they are integrable in the sense of
Definition 3);
(ii) they belong to infinite hierarchies of commuting hydrodynamic chains.
Our method leads to an abundance of new examples of hydrodynamic chains which
require a further detailed investigation. We will address these issues in the future.
We hope that the results of this paper will find important applications in the theory
of infinite-dimensional Frobenius manifolds (yet to be constructed).
Appendix 1: the invariant formulation of the semi-Hamiltonian property
For m-component systems (7) there exists a tensor object which is responsible for the
semi-Hamiltonian property. First of all one computes (1, 3)-tensors M and K ,
p q p q p q
j = Nkp vq Ni j + N pq vk Ni j − N pq Nik v j
s s s s
Mki
p q p q p q
−N pq
s
Nk j vi − Nkp
s
Niq v j − Nkp
s
Nq j vi
and
p p p p p
j = b p ∂u k Ni j − bk ∂u p Ni j + Ni j ∂u p bk − Nkp ∂u i b j + Nkp ∂u j bi
s s s s s s
K ki
p p p p p
+bsp ∂u i N jk − bi ∂u p N sjk + N jk ∂u p bis − Nisp ∂u j bk + Nisp ∂u k b j
p p p p p
+bsp ∂u j Nki − b j ∂u p Nkis + Nki ∂u p bsj − N sj p ∂u k bi + N sj p ∂u i bk ;
123
p
here b = v 2 , that is, bij = v ip v j . Using M and K one defines a (1, 3)-tensor Q as
p q p q p q p q
Q ski j = vk K spq j vi + vk K spiq v j − vq vk K spi j − K kpq
s
vi v j
p p p
+4vk M spi j − 2Mkpj
s
vi − 2Mki
s
pv j .
Ultimately, one introduces a tensor P,

p q p q q p p q
Pkis j = v sp Q kq j vi + v sp Q kiq v j − vqs v p Q ki j − Q skpq vi v j . (58)
Theorem 5 [39] A diagonalizable system (7) is semi-Hamiltonian if and only if the

tensor P vanishes identically.
These objects can be calculated using computer algebra. Notice that they are well-
defined for hydrodynamic chains from the class C: all tensor operations will involve
finite summations only. Invariant coordinate-free definitions of the above tensors can
be found in [39].
It was pointed out in [43, Proposition 5, p. 19] that any strictly hyperbolic conser-
vative system
u it = f i (u)x , i = 1, . . . , m (59)
with the zero Haantjes tensor is necessarily semi-Hamiltonian. The shortest proof of
this statement known to us can be summarized as follows. Let us first rewrite (59) in
the Riemann invariants, Rti = λi (R)Rxi , where λi = λ j due to the strict hyperbolic-
ity. The conserved densities u(R) satisfy an over-determined system of second order
linear PDEs [48],
∂ j λi ∂i λ j
∂i ∂ j u = ∂i u + ∂ j u, i = j. (60)
λ j − λi λi − λ j
The consistency conditions ∂k (∂i ∂ j u) = ∂ j (∂i ∂k u), i = j = k, lead to linear rela-

tions among the first order derivatives of u of the form cn ∂n u = 0 where cn are
certain functions of λ’s. It is well-known that the vanishing of all coefficients cn , that
is, the involutivity of the linear system (60), is equivalent to the semi-Hamiltonian
property (9). In this case we have infinitely many conserved densities parametrized
by m arbitrary functions of a single variable. It turns out that the requirement of the
existence of m functionally independent solutions of the linear system (60) is already
sufficient to conclude that all coefficients cn must be zero. Indeed, if a relation of the
form cn ∂n u = 0 is satisfied by m functionally independent u’s, the vector cn will
be a zero eigenvector of the corresponding Jacobian matrix which is non-degenerate.
Thus, cn = 0 and we have shown that
(m conservation laws) + (Riemann invariants) ⇒ (∞ of conservation laws).
It would be of interest to obtain a direct tensor proof of the above result by showing
that the vanishing of the Haantjes tensor H for a conservative system (59) implies
123
the vanishing of the tensor P. Such a proof would then generalize to hydrodynamic
chains, for which we have yet no rigorous definition of Riemann invariants.
Appendix 2: higher conservation laws for Hamiltonian chains
Let us assume that there exists a conservation law of the form
P(u 1 , u 2 , u 3 , u 4 )t = Q(u 1 , u 2 , u 3 , u 4 , u 5 )x .
Using (13) and collecting coefficients at u ix we obtain the expressions for Q i = ∂ Q/∂u i ,
2 3
Q 1 = 5u 5 P4 h 13 + (4u 4 P5−i h 1,(i+1) + u 1 P3−i h 1,i ) + (3u 3 P5− j h 1, j
i=1 j=1
+2u P4− j h 1, j ) + P1 h 2 ,
2
2 3
i=1 j=1
+2u P4− j h 2, j ) + P2 h 2 + 2P1 h 3 ,

2
2 3
i=1 j=1
+2u P4− j h 3 j ) + P3 h 2 + 2P2 h 3 ,

2
Q 4 = h 2 P4 + 2h 3 P3 ,
Q 5 = 2h 3 P4 .
The consistency conditions Q i j = Q ji imply the following expressions for second

order partial derivatives of P,
6u 4 P4 h 213 + 9u 3 P4 h 12 h 13 + 3u 2 P4 (h 212 + 2h 13 h 11 ) + 3u 1 P4 h 11 h 12 + 2P3 h 3 h 11

P11 = ,
2h 23
12u 4 P4 h 23 h 13 + 9u 3 P4 (h 22 h 13 + h 23 h 12 ) + 6u 2 P4 (h 12 (h 22 + h 13 ) + h 23 h 11 )
P12 =
4h 23
3u 1 P4 (h 212 + h 22 h 11 ) + 2h 3 (3 p4 h 11 + 2 p3 h 12 )
+ ,
4h 23
12u 4 P4 h 13 h 33 + 9u 3 P4 (h 23 h 13 + h 33 h 12 ) + 6u 2 P4 (h 213 + h 23 h 12 + h 33 h 11 )
P13 =
4h 23
3u 1 P4 (h 13 h 12 + h 23 h 11 ) + 2h 3 (3P4 h 12 + 2P3 h 13 )
+ ,
4h 23
3P4 h 13
P14 = ,
2h 3
123
6u 4 P4 h 223 +9u 3 P4 h 22 h 23 +3u 2 P4 (h 222 +2h 23 h 12 )+3u 1 P4 h 12 h 22 +2h 3 (3P4 h 12 + P3 h 22 )

P22 = ,
2h 23
12u 4 P4 h 23 h 33 + 9u 3 P4 (h 223 + h 22 h 33 ) + 6u 2 P4 (h 23 (h 22 + h 13 ) + h 33 h 12 )
P23 =
4h 23
3u 1 P4 (h 13 h 22 + h 23 h 12 ) + 2h 3 (3P4 h 13 + 2P3 h 23 + 3P4 h 22 )
+ ,
4h 23
3P4 h 23
P24 = ,
2h 3
6u 4 P4 h 233 +3P4 (3u 3 h 33 h 23 +u 2 (h 223 +2h 33 h 13 )+u 1 h 23 h 13 )+2h 3 (3P4 h 23 + P3 h 33 )
P33 = ,
2h 23
3P4 h 33
P34 = ,
2h 3
P44 = 0.
Upon ensuring the consistency of this system we obtain the set of relations (29). We
believe that the same process applied to any higher order conservation law
P(u 1 , . . . , u m )t = Q(u 1 , . . . , u m+1 )x
will results in the same relations (29).
Acknowledgments We thank Karima Khusnutdinova for providing us with a program for computing
components of the Haantjes tensor and a further help with computer algebra. We also thank John Gibbons,
Boris Kupershmidt and Maxim Pavlov for their interest and clarifying discussions, and the referee for useful
comments. The research of EVF was partially supported by the European Union through the FP6 Marie
Curie RTN project ENIGMA (Contract number MRTN-CT-2004-5652), the ESF programme MISGAM
and the EPSRC grant EP/D036178/1.
References
1. Alber, S.J.: Associated integrable systems. J. Math. Phys. 32, 916–922 (1991)
2. Benney, D.J.: Some properties of long nonlinear waves. Stud. Appl. Math. 52, 45–50 (1973)
3. Blaszak, M.: Classical R-matrices on Poisson algebras and related dispersionless systems. Phys. Lett.
A 297(3–4), 191–195 (2002)
4. Bukhshtaber, V.M., Leikin, D.V., Pavlov, M.V.: Egorov hydrodynamic chains, the Chazy equation,
and the group SL(2, C). Funktsional Anal. Prilozhen. 37(4), 13–26 (2003)
5. Chazy, J.: Sur les équations différentielles dont l’intégrale générale poss ède un coupure essentielle
mobile. C.R. Acad. Sci. Paris 150, 456–458 (1910)
6. Dubrovin, B.A., Novikov, S.P.: Hydrodynamics of weakly deformed soliton lattices: differential geom-
etry and Hamiltonian theory. Russ. Math. Surv. 44(6), 35–124 (1989)
7. Dubrovin, B.A.: Geometry of 2D topological field theories. Lect. Notes in Math, 1620. Springer,
Heidelberg, 120–348 (1996)
8. Dubrovin, B.A., Liu, S., Zhang, Y.: On Hamiltonian perturbations of hyperbolic systems of conser-
vation laws. I. Quasi-triviality of bi-Hamiltonian perturbations. Comm. Pure Appl. Math. 59(4), 559–
615 (2006)
9. Ferapontov, E.V.: Nonlocal Hamiltonian operators of hydrodynamic type: differential geometry and
applications. Am. Math. Soc. Transl. 170(2), 33–58 (1995)
10. Ferapontov, E.V., Khusnutdinova, K.R.: On integrability of (2+1)-dimensional quasilinear systems.
Comm. Math. Phys. 248, 187–206 (2004). arXiv:nlin.SI/0305044
123
11. Ferapontov, E.V., Khusnutdinova, K.R.: The characterization of two-component (2+1)-dimensional

integrable systems of hydrodynamic type. J. Phys. A Math. Gen. 37(N8), 2949–2963 (2004). arXiv:nlin.
SI/0310021
12. Ferapontov, E.V., Khusnutdinova, K.R.: Hydrodynamic reductions of multi-dimensional dispersionless
PDEs: the test for integrability. J. Math. Phys. 45(N6), 2365–2377 (2004). arXiv:nlin.SI/0312015
13. Ferapontov, E.V., Khusnutdinova, K.R.: Double waves in multi-dimensional systems of hydrodynamic
type: the necessary condition for integrability. Proc. R. Soc. A 462, 1197–1219 (2006). arXiv:nlin.
SI/0412064
14. Ferapontov, E.V., Khusnutdinova, K.R., Pavlov, M.V.: Classification of integrable (2+1)-dimensional
quasilinear hierarchies. Theor. Math. Phys. 144, 35–43 (2005)
15. Ferapontov, E.V., Khusnutdinova, K.R., Marshall, D.G., Pavlov, M.V.: Classification of integrable
Hamiltonian hydrodynamic chains associated with Kupershmidt’s brackets. J. Math. Phys. 47(1) (2006)
16. Ferapontov, E.V., Khusnutdinova, K.R., Tsarev, S.P.: On a class of three-dimensional integrable La-
grangians. Comm. Math. Phys. 261(N1), 225–243 (2006)
17. Gibbons, J.: Collisionless Boltzmann equations and integrable moment equations. Physica D 3, 503–
511 (1981)
18. Gibbons, J., Kodama, Y.: A method for solving the dispersionless KP hierarchy and its exact solutions
II. Phys. Lett. A 135(3), 167–170 (1989)
19. Gibbons, J., Tsarev, S.P.: Reductions of the Benney equations. Phys. Lett. A 211, 19–24 (1996)
20. Gibbons, J., Tsarev, S.P.: Conformal maps and reductions of the Benney equations. Phys. Lett. A 258,
263–271 (1999)
21. Gibbons, J., Raimondo, A.: Differential geometry of hydrodynamic Vlasov equations, arXiv:nlin.SI/
0612022
22. Haantjes, J.: On X m -forming sets of eigenvectors. Indagationes Math. 17, 158–162 (1955)
23. Konopelchenko, B.G., Martinez Alonso, L., Ragnisco, O.: The ∂-approach to the dispersionless KP
hierarchy. J. Phys. A 34(47), 10209–10217 (2001)
24. Krichever, I.M.: The τ -function of the universal Whitham hierarchy, matrix models and topological
field theorie. Comm. Pure Appl. Math. 47(4), 437–475 (1994)
25. Kupershmidt, B.A., Manin, Yu.I.: Long wave equations with a free surface. I. Conservation laws and
solutions (Russian). Funktsional. Anal. Prilozhen. 11(3), 31–42 (1977)
26. Kupershmidt, B.A., Manin, Yu.I.: Long wave equations with a free surface. II. The Hamiltonian struc-
ture and the higher equations (Russian). SFunktsional. Anal. Prilozhen. 12(1), 25–37 (1978)
27. Kupershmidt, B.A.: Deformations of integrable systems. Proc. Roy. Irish Acad. Sect. A 83(1), 45–74
(1983)
28. Kupershmidt, B.A.: The quasiclassical limit of the modified KP hierarchy. J. Phys. A: Math. Gen 23,
871–886 (1990)
29. Kupershmidt, B.A.: Normal and universal forms in integrable hydrodynamical systems. In: Proceed-
ings of the Berkeley–Ames conference on nonlinear problems in control and fluid dynamics (Berkeley,
Calif., 1983), pp 357–378, in Lie Groups: Hist. Frontiers and Appl Ser. B: Systems Inform. Control,
II, Math Sci Press, Brookline (1984)
30. Kupershmidt, B.A.: Extended equations of long waves, preprint (2005)
31. Lebedev, D.R., Manin, Yu.I.: Conservation laws and Lax representation for Benney’s long wave equa-
tions. Phys. Lett. A 74, 154–156 (1979)
32. Mañas, M.: On the r th dispersionless Toda hierarchy: factorization problem, additional symmetries
and some solutions. J. Phys. A Math. Gen. 37, 9195–9224 (2004)
33. Mañas, M.: S-functions, reductions and hodograph solutions of the r th dispersionless modified KP
and Dym hierarchies. J. Phys. A Math. Gen. 37, 11191–11221 (2004)
34. Mañas, M., Martinez Alonso, L., Medina, E.: Reductions and hodograph solutions of the dispersionless
KP hierarchy. J. Phys. A Math. Gen. 35, 401–417 (2002)
35. Mikhalev, V.G.: Hamiltonian formalism of Korteweg-de Vries-type hierarchies (Russian). Funktsional.
Anal. i Prilozhen. 26(2) 79–82 (1992); translation in Funct. Anal. Appl. 26(2) 140–142 (1992)
36. Miura, R.M.: Conservation laws for fully nonlinear long wave equations. Stud. Appl. Math. 53,
45–56 (1974)
37. Pavlov, M.V.: Integrable hydrodynamic chains. J. Math. Phys. 44(9), 4134–4156 (2003)
38. Pavlov, M.V.: Classification of integrable Egorov hydrodynamic chains. Teoret. Mat. Fiz. 138(1),
55–70 (2004)
123
39. Pavlov, M.V., Svinolupov, S.I., Sharipov, R.A.: An invariant criterion for hydrodynamic integrability.
Funktsional Anal. Prilozhen. 30,18–29 (1996). Translation in Funct Anal Appl 301522
40. Pavlov, M.V.: Modified dispersionless Veselov–Novikov equations and corresponding hydrodynamic
chains, arXiv:nlin.SI/0611022
41. Pavlov, M.V.: The Hamiltonian approach in classification and integrability of hydrodynamic chains,
arXiv:nlin.SI/0603057
42. Peradzyński, Z.: Nonlinear plane k-waves and Riemann invariants. Bull. Acad. Polon. Sci. Sr. Sci. Tech.
19, 625–632 (1971)
43. Sévennec, B.: Géométrie des systèmes hyperboliques de lois de conservation, Mémoire (nouvelle série)
N56, Supplément au Bulletin de la Société Mathématique de France 122, 1–125 (1994)
44. Martinez Alonso, L., Shabat, A.B.: Towards a theory of differential constraints of a hydrodynamic
hierarchy. J. Nonlinear Math. Phys. 10(2), 229–242 (2003)
45. Martinez Alonso, L., Shabat, A.B.: Energy-dependent potentials revisited: a universal hierarchy of
hydrodynamic type. Phys. lett. A 299(4), 359–365 (2002); Phys. Lett. A 300(1), 58–64 (2002)
46. Teshukov, V.M.: On the hyperbolicity of long wave equations. Dokl. Akad. Nauk SSSR 284(3), 555–
559 (1985)
47. Teshukov, V.M.: Characteristics, conservation laws and symmetries of the kinetic equations of motion
of bubbles in a fluid. Prikl. Mekh. Tekhn. Fiz. 40(2), (1999) 86–100; translation in J. Appl. Mech.
Tech. Phys. 40(2), 263–275 (1999)
48. Tsarev, S.P.: Geometry of Hamiltonian systems of hydrodynamic type. Generalized hodograph method.
Izvestija AN USSR Math. 54(5), 1048–1068 (1990)
49. Zakharov, V.E.: Benney equations and quasiclassical approximation in the inverse problem
method. Funktsional Anal. i Prilozhen. 14(2), 15–24 (1980)
50. Zakharov, V.E.: On the Benney equations. Physica D 3(1,2), 193–202 (1981)
123

Hantjes Tensor

Uploaded by

Copyright:

Available Formats

Hantjes Tensor

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hantjes Tensor

Uploaded by

Copyright:

Available Formats

Math. Ann.

Differential-geometric approach to the integrability

Abstract The integrability of an m-component system of hydrodynamic type, ut =

Mathematics Subject Classification (2000) 35L40 · 35L65 · 37K10

Hydrodynamic chains are quasilinear first order PDEs of the form

E. V. Ferapontov (B) · D. G. Marshall

where u = (u 1 , u 2 , . . .)t is an infinite-component column vector and V (u) is an ∞×∞

obtained in [28] as a quasiclassical limit of the modified KP hierarchy. Its two-param-

naturally appears in the theory of finite-gap solutions of integrable hierarchies of the

u 1t = u 2x , u 2t = g(u 1 , u 2 , u 3 )x , u 3t = h(u 1 , u 2 , u 3 , u 4 )x , . . . , (6)

Our approach to the integrability of hydrodynamic chains is motivated by the theory

Rti = λi (R)Rxi (8)

∂k = ∂/∂ R k , i = j = k. We emphasize that the semi-Hamiltonian property (9)

and introduce the Haantjes tensor

Theorem 1 [22] A hydrodynamic type system with mutually distinct characteristic

the compatibility conditions). The requirement of the vanishing of other components

the right hand side of the fourth equation u t = . . ., etc.

The characterization of diagonalizable chains of a more general form,

u 1t = f (u 1 , u 2 )x , u 2t = g(u 1 , u 2 , u 3 )x , u 3t = h(u 1 , u 2 , u 3 , u 4 )x , . . . , (12)

here B i j = (i − 1)u i+ j−2 and h(u 1 , u 2 , u 3 ) is a Hamiltonian density. The Benney

In Sect. 5 we prove our main result:

Theorem 2 provides an easy-to-verify necessary condition for testing the integra-

2 Classification of diagonalizable chains of the type (6)

of the ten second order partial derivatives of h(u 1 , u 2 , u 3 , u 4 ):

2g1 g12 − g2 g11 + 2h 1 g13

Notice that these equations can be compactly written as

the consistency conditions pt y = p yt are satisfied identically modulo the dis-

2.1 Integration of the system (15)

To explicitly calculate g(u 1 , u 2 , u 3 ) we will follow [38]. The main observation is

Here η(u 1 ) is a solution to the Chazy equation [5],

η + 2ηη − 3η = 0,

and σ (u 1 , u 2 ) is an elliptic sigma function in the variable u 2 whose dependence on

The last three equations imply

The last three equations imply

Up to a transformation of the form u 3 → u 3 + αu 2 + βu 1 + γ this implies (19). It was

The last three equations imply

Rewritten in terms of v, the third equation for p implies

v22 = 6v 2 − 4vη − 4η , (23)

The condition of their consistency leads to the Chazy equation for η.

3 Classification of diagonalizable chains of the type (12)

h 4 g13 h 4 g23 h 4 g33

f 11 (−g33 g2 g1 +g3 (g23 g1 +g2 g13 −g3 g12 ))+2 f 12 (g13

f 22 f 11 g32 − f 2 ( f 1 f 11 g33 + f 11 (g3 g23 − g33 g2 ) + 2 f 12 (g33 g1 − g3 g13 ))

3.1 Integration of the Eqs. (26), (27)

p12 (2 p12 − f 11 ) + 2 p11 ( f 12 − p22 )

f 11 ( f 12 − p22 ) + 2 f 12 p12 − f 22 p11

a, b, A, B=const. The substitution of these expressions into the remaining equations

The case a = 1, b = 0 leads to

p22 + f 12 + f 2 p1 − 4 p2 f 1 ( f 1 − p2 )( f 11 − 4 p12 ) + 4 p1 ( f 12 − p22 )

2 p12 p1 (4 p12 − f 11 ) + 4 p11 ( f 1 − p2 )

f = s(u 1 )(u 2 )k , p = q(u 1 )(u 2 )k+1 ,

f = (au 1 + b) ln u 2 + s(u 1 ), p = q(u 1 )u 2 + a(u 2 ln u 2 − u 2 ),

p22 (2 p12 −2 p11 )+ p11 ( p2 (2 p2 −4 f 1 )+2( f 12 + f 12 + f 2 p1 ))+ p12 (2 p12 − f 11 +4 p1 ( f 1 − p2 ))

Although this system is in involution and reduces to the corresponding Case 4 in

4 Classification of diagonalizable Hamiltonain chains

where B i j = (i − 1)u i+ j−2 , and h = (u 3 + (u 1 )2 )/2 is the Hamiltonian density.

p111 (2 + u 1 p22 ) = p11 p22 − p12

which, up to a natural equivalence h → αh + au 1 + bu 2 + c, lead to the Hamiltonian

h = u 3 + α(u 1 )2 + βu 1 u 2 + γ (u 2 )2 + δ(u 1 )3 (28)

η + 2ηη − 3η = 0,

v22 = 6v 2 − 4vη − 4η , (23)

4c2 η2 g − 12c2 ηg − 1 − αη + 2cη2 = 0,

(4c2 η2 g − 12c2 ηg − 1 − αη + 2cη2 )g + (α − 4cη + 12c2 g + 4c2 ηg

(4c2 η2 g − 12c2 ηg − 1 − αη + 2cη2 )g = 0.

8z 2 s s + 8zs s − 4z 2 (s )2 − (s )2 − 12s = 0

The semi-Hamiltonian property is satisfied if and only if p = 0. Thus, we have con-