Unified structures for solutions of Painlevé equation II.

Federico Zullo

DICATAM, Università degli Studi di Brescia via Branze, 38 - 25123 Brescia, Italy
& INFN, Milano Bicocca, Piazza della Scienza 3, Milano, 20126, Italy
e-mail: [email protected]
arXiv:2412.21018v1 [nlin.SI] 30 Dec 2024

Abstract
We present certain general structures related to the solutions of Painlevé equation II and to
the solutions of the differential equation satisfied by the corresponding Hamiltonian equa-
tions, together with the tau functions. By taking advantage of the Bäcklund transformations
we find a linear differential equation satisfied by the tau functions and explicit rational ex-
pressions linking the solutions of Painlevé equation II, Painlevé equation XXXIV and the
Hamiltonians with the tau functions. Also, recursions for the tau functions and the Hamilto-
nians will be given. Through the use of suitable parameters inserted into the equations, we
consider a degenerate case solvable in terms of Weierstrass elliptic functions. The Yablonskii-
Vorob’ev polynomials, the Airy-type solutions and the more general transcendental case are
all described in separated Sections. Finally, certain integral relations are given.

KEYWORDS: Painlevé II, Weierstrass elliptic functions, Yablonskii-Vorob’ev polynomials, Airy

functions, tau functions.

1 Introduction
The Painlevé II equation is usually written in the following form:

d2 y
= 2y 3 + zy + α, (1)
dz 2
where α is a parameter. It is well known that this equation possesses, for integer and half-integer
values of α, rational solution and solution explicitly expressible in terms of Airy functions.
All other solutions are transcendental. One of the differences among these solutions is in the
distribution of the zeros in the complex plane: they are finite in number for rational solutions,
are infinitely many in the other two cases but the order of growth of the solutions are different:
it is 3/2 for the Airy type solutions and 3 for the other. If instead of the distribution of the zeros
one takes into account, for example, the asymptotic behaviour of the solutions of (1) on the
real axis, then another variety of solutions with different analytical properties are obtained also
in the case of a fixed value of the parameter α (see e.g. [3] and references therein). This wide
variety of type of solutions prevented the development of a unified notation for the Painlevé
transcendents. A unifying trait for the solutions of (1) is that the poles are all simple with
residues plus or minus 1. Indeed, all the solutions of equation (1) can be represented as the
difference of two functions having simple poles with residues +1: these functions, usually called
σ function [1], from a mechanical point of view are Hamiltonian function and for this reason are
also symbolized with an H (see e.g. [10, 11]). The Painlevé equations are indeed Hamiltonian,

1
i.e. they possess an Hamiltonian structure; the corresponding Hamilton equations correspond
to the Painlevé equations. It is the Hamiltonian, as a function of z, that indeed appears more
frequently in the applications.
In this paper, following [7], we introduce some parameters in the Hamilton equations also
to elucidate the connections between the Painlevé equation II and a degenerate case involving
the Weierstrass elliptic functions. The linking turns out to be not an accident but it reflects
some general structures of the solutions of the Painlevé equation II and and of the corresponding
tau functions that we are going to report and discuss. Actually, these general structures reflect
the common Hamiltonian structure and the common existence of canonical transformations,
depending on the parameters, that represent Bäcklund transformations among different solutions
of the Painlevé equation (apart the elliptic case when one has auto-Bäcklund transformations).
The paper is organized as follows: in Section (2) we introduce a set of parametric Hamiltonians
together with the corresponding canonical variables related to the Painlevé equation (1). Also,
we introduce the Bäcklund transformations, essentially as given by Okamoto [11]. In Section
(3) we get a bilinear equation satisfied by the Hamiltonian functions and then two explicit third
order recurrences. The compatibility between these two gives a second order quadratic recursion.
A discussion will be made on its solutions. In Section (4) we will show how all the solutions
of the main differential equation of this work, i.e. equation (7), can be formally reduced to
the solutions of Riccati equations whose coefficients are certain combinations of Hamiltonians
themselves. By introducing the tau functions corresponding to the Hamiltonians one gets a linear
equation for the tau functions. This linear equation, together with the Toda equation, lead to
explicit expressions of the solutions of the Painlevé equation II, Painlevé equation XXXIV and
the Hamiltonians in terms of rational combinations of tau functions. Also, a fifth order explicit
recursion for the tau functions will be given. In Section (5) a degenerate case, solvable in terms
of elliptic functions, will be investigated. The Bäcklund transformations become auto-Bäcklund
transformations and some explicit formulae, related to addition formulae for the Weierstrass
elliptic functions, will be discussed. In Section (6) the rational solutions of Painlevé II will
be considered in the ligth of the results obtained in Section (4). Some of the results given
in [2] will be re-obtained. In Section (7) the Airy type solutions of the Painlevé II equation
are considered. Again, we will give different relations between the Painlevé II functions, tau
functions and Hamiltonians by specifying the results of Section (4) to this case. In Section (8)
the general transcendental case will be considered. The action of the Bäcklund transformations
on the Taylor series coefficients of solutions regular ta the origin will be investigated. It will be
shown how this action can be described in terms of Weierstrass elliptic functions, generalizing
the results given in Section (5). Finally, in Section (9) some general integral relations, specifying
the action of the Bäcklund transformations on the tau functions, will be presented. Again, these
relations are general and apply to any solution of the Painlevé II equation and the corresponding
Hamiltonian and tau functions.

2 Canonical transformations
We take the following set of Hamiltonian functions, labelled by the discrete index n ∈ Z, written
in terms of the canonical variable (qn , pn ) as:

p2n mg2
Hn = − apn (qn2 + bz + c) − en qn − (2)
2m 2a2
The last term, proportional to g2 , is just a constant and it could be removed by seting it equal
to zero, but we will keep it since it will be useful in the rest of the paper. The other constants
are m, a, b, c, en , with en = e + 2abmn, n ∈ Z, a and m in general different from zero. The

2
corresponding Hamilton equations are given by
dqn pn
= − a(qn2 + bz + c)
dz m (3)
dpn
= en + 2apn qn
dz
The differential equations satisfied by qn and pn are respectively given by

d2 qn en
= 2a2 qn3 + 2a2 (bz + c) qn + − ab,
dz 2 m
(4)
d2 pn 1 dpn 2 2a 3 e2
 
pn 2 = + pn − 2a2 (bz + c)p2n − n .
dz 2 dz m 2

The first equation is the Painlevé II, whereas the second one is known as Painlevé XXXIV (in
the Ince’s notation [9]). It is clear that we could get rid of various constants, for example a
and m, just by rescaling the dependent and independnet variables. In the application however,
for example in the scaling reductions of partial differential equations, usually the independent
variable is a space or time variables or a precise combination of both and through the scaling
one obtains equation (1) only after other reductions. This is one of the reasons we find useful
to leave the parameters in equations (2) and (3).
The derivative of Hn as a function of z is given by
dHn ∂Hn
= = −abpn (5)
dz ∂z
and from (3) it follows that Hn solves the following second order differential equation
2 3 2
d2 Hn
    
2 dHn (z) 2 dHn 2 dHn dHn dHn (z) 2 2 2
+ +4a c +4a b z − Hn −2bmg2 −a b en = 0.
dz 2 bm dz dz dz dz dz
(6)
Equation (6) possesses meromorphic solutions [6]. It possesses rational solutions, solutions
that can be write in terms of elliptic functions (and the corresponding degenerate case given
in terms of trigonometric functions), solutions written in terms of Airy functions and their
derivatives and transcendental solutions with order of growth ≤ 3. We are not aware of any
proof that the transcendental solutions of (6), apart the rational, elliptic and the Airy type
solutions, have order of growth exactly 3 when b 6= 0. When b = 0 all the solutions have
order of growth 2 (or 1, in the trigonemoetric degenerate case) since the general solution can be
reduced to Weierstrass elliptic functions (see Section (5)). Anyway, the non-rational solutions
of (6) possess infinitely many poles of order 1. Also, it is easy to see that all the poles have the
residues equal to 2bm, so we let Hn (z) = 2bmhn (z) with hn (z) having all poles with residue 1.
From now on for simplicity we will denote the derivative with a dot on the dependent variable.
The corresponding differential equation for hn reads

a2 e2n
(ḧn )2 + 4(ḣn )3 + 4a2 (bz + c)(ḣn )2 − (4a2 bhn + g2 )ḣn − = 0. (7)
4m2
This is our main differential equation. When b 6= 0 a simple shift will eliminate the dependence of
hn on g2 . However, as explained above, we want to keep the constant g2 in (7). Notwithstanding,
for b 6= 0, it will be useful to introduce the function kn related to hn by
g2
hn (z) = − + kn (z). (8)
4a2 b

3
 Now let us now consider the following set of canonical transformations:
en
qn+1 = −qn − ,
apn (9)
2
pn+1 = −pn + 2amqn+1 + 2am(bz + c)

The function Hn (pn , qn , z), and hence hn (pn , qn , z), is preserved by the transformations, i.e.

Hn (pn (pn+1 , qn+1 ), qn (pn+1 , qn+1 ), z) = Hn (pn+1 , qn+1 , z).

Also, the generating function of the non-autonomous canonical transformations (9) is given by
en 2am 3
Fn = − ln(qn + qn+1 ) − q − 2ma (c + bz) qn+1 (10)
a 3 n+1
Since the Hamiltonian system defined by the Hamiltonian (2) is non autonomous, it follows that
(qn+1 , pn+1 ), as a function of z, are the integral curves of the new Hamiltonian [18]
∂F
Hn (pn+1 , qn+1 , z) + = Hn (pn+1 , qn+1 , z) − 2mabqn+1 = Hn+1 (pn+1 , qn+1 , z) (11)
∂z
Now we are going to refer all to the function hn and not to Hn . From (11) it follows that the
function qn+1 is proportional to the difference hn − hn+1 :

hn − hn+1 = aqn+1 . (12)

The equations of motion (3) and the canonical transformations (9) gives:
ṗn en
= + 2aqn = a(qn − qn+1 ). (13)
pn pn
The previous, together with (9), gives

ṗn ḧn
a(qn − qn+1 ) = =
pn ḣn (14)
en aen
a(qn + qn+1 ) = − =
pn 2mḣn
Equations (12) and (14) are still valid if one substitutes hn with kn given in equation (8). From
equations (14) it follows:

aen ḧn
aqn = + , (15)
4mḣn 2ḣn
aen ḧn
aqn+1 = − , (16)
4mḣn 2ḣn
and by evaluating the first equation in n → n + 1 and subtracting yields:
   
2m ḧn+1 ḣn + ḣn+1 ḧn+1 + a en+1 ḣn − en ḣn+1 = 0. (17)

The previous can be integrated once, giving:

2mḣn+1 ḣn + a (en+1 hn − en hn+1 ) = C1 , (18)

where, as we will now show, the constant on the right hand side must be equal to −mg2 /2.
Indeed, equations (14) implies
en + ṗn
aqn+1 = − , (19)
2pn

4
whereas from (12) and (19) we get

2mḧn − aen
hn+1 = hn + . (20)
4mḣn

Equation (20) can be considered the Bäcklund transformation for the Hamiltonian hn : if hn is
a solution of (7), then hn+1 given by (20) is a solution of (7) evaluated in n → n + 1. This
can be checked also by inserting equation (20) in the equation (7) evaluated in n → n + 1 and
considering equation (7) itself and its differential consequences. Now we can insert hn+1 from
(20) into (18): we get a third order equation for hn and the compatibility with (7) implies
C1 = −mg2 /2. We also notice that equation (18) written in terms of the shifted hamiltonians
kn (8) is
2mk̇n+1 k̇n + a (en+1 kn − en kn+1 ) = 0. (21)

Now it is possible to get the corresponding inverse transformation of (20) by looking at the
inverse transformations of (9), given by
en
qn+1 = −qn − ,
apn+1 (22)
pn+1 = −pn + 2amqn2 + 2am(bz + c)

By repeating the previous discussion to the inverse transformations (9) we can write the inverse
of the Bäcklund transformations for hn (20) as:

2mḧn+1 + aen+1
hn = hn+1 + . (23)
4mḣn+1

We notice that this equation can be derived also from (15): by evaluating it in n → n + 1 and
using (12) one gets (23). The transformation (20) gives a solution of the equation (7) evaluated
in n → n + 1 from a solution of equation (7) itself. The transformation (23) pull back this
solution. These two transformations define a discrete dynamics over n: in the next Section
we will show how both (20) and (23) provide certain recurrences for the functions hn and the
corresponding tau functions.

3 Recursions for the Hamiltonian functions

It is possible to get another bilinear equation involving the Hamiltonian functions hn and its
derivatives. Adding (20) and (23) we get

ḧn+1 ḧn a
(hn+1 − hn )(ḣn+1 − ḣn ) + + + (en+1 − en ) = 0, (24)
2 2 4m
and, by taking into account also that en+1 − en = 2abm and integrating once

(hn+1 − hn )2 + ḣn+1 + ḣn + a2 bz = C2 , (25)

By inserting (20) in (25) and using (7) we see that the constant C2 must be equal to −a2 c. So
we get
(hn+1 − hn )2 + ḣn+1 + ḣn + a2 (bz + c) = 0. (26)
The previous equation can be seen as a Riccati equation for hn+1 or for hn : the coefficients of the
Riccati equation involve the adjacent Hamiltonian. We will return on this equation in Section

5
(3). Now we will obtain an explicit expression for the derivative of hn in terms of Hamiltonian
functions: this will give two explicit recursions for hn from equations (26) and (18).
By evaluating (23) in n → n − 1 and subtracting with (20) we get the following explicit
expression for ḣn :
aen
ḣn = − . (27)
2m(hn+1 − hn−1 )
It is possible to use the previous equation, together with (26) and (18) to obtain an explicit
recursion for the function hn+3 (z) in terms of hn+2 (z), hn+1 (z) and hn (z). From (26) we have
aen+2 aen+1
(hn+2 − hn+1 )2 − − + a2 (bz + c) = 0, (28)
2m(hn+3 − hn+1 ) 2m(hn+2 − hn )
that is, by isolating hn+3 :
aen+2 hn+2 − hn
hn+3 = hn+1 + aen+1 . (29)
2m (hn+2 − hn ) (a (bz + c) + (hn+2 − hn+1 )2 ) −
2
2m

Equation (18) gives

aen+2 en+1 mg2
+(hn+3 −hn+1 )(hn+2 −hn )(hn+1 en+2 −hn+2 en+1 )+ (hn+3 −hn+1 )(hn+2 −hn ) = 0,
2m 2a
(30)
and, by isolating again hn+3 we get:

a2 en+1 en+2
hn+3 = hn+1 + (31)
m(hn+2 − hn )(mg2 + 2a(hn+1 en+2 − hn+2 en+1 ))

The compatibility between equation (29) and (29) gives the following second order equation for
hn+2 :
m
(hn+2 − hn )2 4a2 bhn+1 + g2 +


(hn+2 − hn+1 )(hn+2 − hn )(hn+1 − hn ) −
2aen+1
aen+1 (32)
− a2 (bz + c) (hn+2 − hn ) + = 0.
2m
Let us make some comments on the above recurrences. The dependence on g2 disappears (as it
must be) from (30) and (31) if we make the shift (8). One has
aen+2 en+1
+ (kn+3 − kn+1 )(kn+2 − kn )(kn+1 en+2 − kn+2 en+1 ) = 0, (33)
2m
and
aen+1 en+2 1
kn+3 = kn+1 + (34)
2m (kn+2 − kn )(kn+1 en+2 − kn+2 en+1 )
Equations (28) and (29) remain unaltered by the shift instead. The recursion (32) becomes
2abm
(kn+2 − kn+1 )(kn+2 − kn )(kn+1 − kn ) − (kn+2 − kn )2 kn+1 +
en+1 (35)
aen+1
− a2 (bz + c) (kn+2 − kn ) + = 0.
2m

A second comment: the difference equation (32), or (35), is quadratic in the three variables
hn , hn+1 and hn+2 (or kn , kn+1 and kn+2 ). It is known that the Bäclund transformations (9),
and hence (32), produce rational solution from rational solutions and special solutions written as
rational combinations of Airy functions from the Airy-type solutions. This means that, at least
in these cases, the discriminant of equation (32) with respect to one of the variables, say hn+2 , is

6
a perfect square. Actually we will now show that this discriminant is always a perfect square and
this is very useful in identifying the solutions of equation (32) and hence the explicit recursion
for hn+2 in terms of the other variables. Clearly this should be consistent with equations (29)
and (31). At first sight it seems that (32) is different from (29) and (31), in the sense that (32)
is a second order difference equation, whereas (29) and (31) are third order equations. Actually
we will show that a solution of (32) is related to hn−1 . The main reason for this is the expression
(27) relating the derivative of hn to hn−1 .
The discriminant ∆n of equation (32) with respect to hn+2 is

2 2aen+1
∆n = (hn+1 − hn )2 + a2 (bz + c) + 4a2 bhn+1 − (hn+1 − hn ) + g2 . (36)
m
In order to identify if the previous can be written as a perfect square we use the identity (20)
for hn+1 in terms of hn . By carefully looking at the corresponding equation for ∆n we arrive at
the following expression:
 2 2
 2mḧn − aen + 16a2 m2 (bz + c)ḣ2n + 2 3
32m ḣn 
∆n =   . (37)
16m2 ḣ2n

Indeed, the difference between (36) and (37) is proportional to the left hand side of equation (7)
and therefore is zero. By taking into account the expression (37), we get two explicit values for
the roots of equation (32) with respect to hn+2 . Let us call them x+ and x− . By choosing the
sign + for the square root of (37) and by using the expression (20) and the differential equation
(7) to eliminate the even powers of ḧn we get
aen+1 en+1
x+ = hn − = hn + (hn+1 − hn−1 ) . (38)
2mḣn en

Whereas, by choosing the sign − for the square root of (37) and again by using the expression
(20) and the differential equation (7) to eliminate the even powers of ḧn we get

4aen+1 mḣ2n+1
x− = hn − . (39)
2men ḧn + 2m2 (4a2 bhn + g2 )ḣn + a2 e2n

The root x− gives hn+2 . Indeed the expression (39) can be obtained by the repeated use of the
Bäcklund transformation (20), i.e. by shifting the n in (20) to n + 1 and then using (20) again
to get a formula relating hn+2 to hn . By doing this one obtains a huge formula for hn+2 in terms
of hn and its derivatives up to the fourth order, but this fromula can be simplied by taking into
account the differential equation (7) and its differential consequences. What we get is exactly
formula (39), i.e. x− = hn+2 .
So we found the two roots of the quadratic recurrence (32): one is hn+2 , the other is given
by x+ in (38). At this point we can get explicit formulae relating hn+2 to hn+1 , hn and hn−1 .
Indeed, the two polynomials in x
m
(x − hn )2 4a2 bhn+1 + g2 +


(x − hn+1 )(x − hn )(hn+1 − hn ) −
2aen+1
aen+1 (40)
− a2 (bz + c) (x − hn ) + = 0.
2m
and
m(4a2 bhn+1 + g2 )
  
en+1
hn+1 − hn − x − hn − (hn+1 − hn−1 ) (x − hn+2 ) (41)
2aen+1 en

7
must have the some roots. By equating the coefficients of x and x0 one gets two expressions
for hn+2 in terms of hn+1 , hn and hn−1 . These two expressions are indeed compatible with (29)
and (31) if one takes into account equation (32) itself.
As far as we know both the discrete equation (32) satisfied by the Hamilton functions of the
Painlevé equation II and the explicit solutions (29) and (31) are new. Once again, we stress
that the first element of the chain h0 could be a rational solution, a special type solution or
a transcendental solution. The successive Hamiltonians can be obtained by means of (20) and
satisfy (29), (31) and (32)
In the next Section we consider further equation (26) and will give a linear equation satisfied
by the corresponding tau functions.

4 Linearization of the solutions of the Painlevé II equation

It is well known that special solutions of the Painlevé II equation can be obtained as solution
of a Riccati equation. Indeed, it is the function hn (z) that satisfies a Riccati equation and
the corresponding solutions of the Painlevé II are obtained through the relation (12). For
completeness, we remember how the special solutions of Painlevé II are obtained. We look for
solutions of equation (7) that are also solutions of a Riccati equation, i.e.

ḣn = Ah2n + Bhn + C, (42)

where A, B and C are some indeterminate functions. By taking the derivative of (42), inserting
it in (7) and taking into account (42) itself we get a sixth order polynomial for hn . We look for
conditions on the functions A, B and C so that all the coefficients of the powers of hn vanish.
The coefficients of h6n gives A = −1. By inserting this value into the sixth order polynomial we
get a fourth order polynomial in hn . The coefficient of h4n is given by

B2
C+ + Ḃ + a2 (bz + c) (43)
4
and fixes the value of C in terms of B. By inserting this into the fourth order polynomial we
get a third order polynomial whose leading coefficient is given by B̈. So we set

B = α + βz. (44)

With this value of B we get a second order polynomial in hn . The vanishing of the first and
second order coefficients fix β = 0 and α = − 2ag22 b , whereas the order zero coefficients gives the
condition e2n = 4a2 b2 m2 . By remembering that en = e + 2abmn, it follows e = 0 and n = ±1.
So, if e = 0 and n = ±1, equation (7) can be reduced to
g2
ḣn + h2n + (8a2 bhn + g2 ) + a2 (bz + c) (45)
16b2 a4
and, by letting hn = − 4ag22 b + τ̇n
τn , one has

τ̈n = −a2 (bz + c)τn , (46)

i.e. τn is an Airy function. Through the use of the Bäcklund transformations (20) and (23) one
gets the family of special solutions corresponding to e = 0 and n ∈ N in (7).
Now we will show that actually all the solutions of the equation (7) can be formally reduced
to a solutions of a Riccati equation. The point is that the functions A, B and C in equation
(42) could depend on a set of functions hn+m for some subset of integers m and equation (42)

8
could be satisfied due to some of the recurrences found in the above Sections. Let us start by
noticing that, by evaluating (20) in n → n + 1 and subtracting with (23) it follows
aen+1
hn+2 − hn = − . (47)
2mḣn+1
Differentiating we get
aen+1 ḧn+1
ḣn+2 − ḣn = . (48)
2m(ḣn+1 )2
Now we look at the right hand side of (48). By equation (27) it follows
aen+1
ḣn+1 = − , (49)
2m(hn+2 − hn )

and, from (23) and (49) we get

ḧn+1
= hn+2 − 2hn+1 + hn . (50)
ḣn+1

Equations (48), (49) and (50) together give

ḣn+2 = ḣn − (hn+2 − 2hn+1 + hn ) (hn+2 − hn ) . (51)

The previous is indeed a Riccati equation for hn+2 , with the coefficients A, B and C in (42)
explicitly given by

A = −1, B = 2hn+1 , C = ḣn + h2n − 2hn hn+1 . (52)

Notice that hn+2 (z) is an arbitrary solution of equation (7) evaluated in n → n + 2, the constant
en+2 being e + 2amb(n + 2) with e arbitrary.
Equation (51) can be obtained also directly from (26) by evaluating it in n → n + 1 and
subtracting (26) itself.
Let us introduce the τn function through the logarithmic derivative
τ̇n
hn (z) = . (53)
τn
The functions τn are entire and possess simple zeros. Notice that the function τn+2 solves a
second order linear equation with coefficients depending on different τ functions. Explicitly we
have
τn+1 (τ̈n+2 τn − τ̈n τn+2 ) − 2τ̇n+1 (τ̇n+2 τn − τ̇n τn+2 ) = 0. (54)
The previous can be integrated once, giving
2
τ̇n+2 τn − τ̇n τn+2 = αn τn+1 , (55)

where αn is a constant that in principle could also depend on n. Notice that equation (55) is
linear in τn and τn+2 . Together with equation (55) we should also consider the so called Toda
equation, that can be obtained from (50). Indeed inserting the definitions (53) in (50) and
integrating once we get
2
τ̈n+1 τn+1 − τ̇n+1 = βn τn τn+2 , (56)
where again βn is a constant of integration. The constant αn and βn actually depend on the
gauge chosen to normalize each τn function: from (53) we see that indeed each τn is defined
up to a scaling factor. We could fix it for example by taking the first non-zero Taylor series

9
coefficients of τn in z = 0 to be equal to 1. So, for example, if z = 0 is not a zero of τn we can
fix τn (z) = 1 + O(z) around z = 0. However, from equation (49) we see that
aen+1
αn βn = − (57)
2m
implying that, in general, they depend on n and are both different from zero. Apart (55) and
(56), other two differential equations can be obtained for the tau functions. They come directly
from equations (18) and (26) and the use of (56). We get:
a g2
(en τn τ̇n+1 − en+1 τn+1 τ̇n ) = βn βn−1 τn+2 τn−1 + τn τn+1 , (58)
2m 4
and
τ̈n+1 τn + τ̈n τn+1 − 2τ̇n+1 τ̇n + a2 (bz + c)τn τn+1 = 0. (59)
This last equation can be cast also in the following form:
 2
τ̇n+1 τ̇n .
ÿn + − + a2 (bz + c) = 0, yn = ln(τn τn+1 ) (60)
τn+1 τn

and, by introducing the Wronskian between two successive tau functions, i.e.

Wn (z) = τ̇n+1 τn − τn+1 τ̇n = (hn+1 − hn )τn τn+1 , (61)

we see that it can be rewritten as

 2
Wn
ÿn + + a2 (bz + c) = 0. (62)
τn+1 τn

Also, if we use (56) we get:

 2
Wn βn τn τn+2 βn−1 τn−1 τn+1
+ 2 + + a2 (bz + c) = 0 (63)
τn+1 τn τn+1 τn2

Actually we can get an explicit formula for Wn itself (not its square) in terms of tau functions.
Indeed, let us take the derivative of equation (55) shifted in n to n − 1 and use equation (56)
and its suitable shift in n to eliminate the second derivatives. We get:
2 τ2 2 2
τn−1 τn τn+2 τn−2 τn τn+1 τ̇n+1 n−1 − τ̇n−1 τn+1
βn − βn−2 + = 2αn−1 τn τ̇n . (64)
τn+1 τn−1 τn+1 τn−1

We use again (55) shifted in n → n − 1 in the previous. We get:

τn−1 τn τn+2 τn−2 τn τn+1 αn−1 τn
βn − βn−2 = (2τ̇n τn+1 τn−1 − τ̇n+1 τn τn−1 − τ̇n−1 τn τn+1 ) . (65)
τn+1 τn−1 τn+1 τn−1

The previous can be rewritten in terms of Wronskians as

2 2
βn τn−1 τn+2 − βn−2 τn−2 τn+1 = αn−1 (τn+1 Wn−1 − τn−1 Wn ) . (66)

Further, we notice that, due to equation (55), it holds another linear relation between Wn and
its neighbours, i.e.
τ3 τn−1
Wn−1 = αn−1 n − Wn . (67)
τn+1 τn+1

10
Equations (67) and (66) can be solved for Wn and Wn−1 giving
2 τ 2
αn−1 τn3 βn−2 τn+1 n−2 − βn τn−1 τn+2
Wn = + , (68)
2 τn−1 2αn−1 τn−1
2 τ 2
αn−1 τn3 βn−2 τn+1 n−2 − βn τn−1 τn+2
Wn−1 = − , (69)
2 τn+1 2αn−1 τn−1

The previous two equations are compatible with (63). Indeed, if we consider the difference
between (63) evaluated in n → n + 1 and itself, then by using (67) we arrive at formula (68)
and (69) again.
Equations (68) and (69) are quite remarkable: indeed the left hand side are entire functions
and so are the right hand side. It follows that the zeros of the denominator correspond to zeros
of the numerator. Further, equations (68) and (69) give an explicit recursion for τn+5 in terms
of the four previous tau functions:

aen+4 α2 αn+3 τn+3

3 2 τ
a2 bαn+3 τn+4 2
n+1 αn+1 αn+2 αn+3 τn+2 τn+4 aen+1 αn+2 αn+3 τn τn+3 τn+4
τn+5 = n+2 2 − 2 − + 2
2m τn+2 αn+1 τn+2 τn+1 2m αn αn+1 τn+1 τn+2
(70)
We end this Section by giving two equivalent expressions of the solution qn of the Painlevé
equation II and of the Painlevé equation XXXIV as a rational combination of tau functions.
From equations (12), (68) and (69) we get
2 τ 2
αn−1 τn2 βn−2 τn+1 n−2 − βn τn−1 τn+2
aqn = − + , (71)
2 τn+1 τn−1 2αn−1 τn−1 τn τn+1

and
2 2 τ 2
αn−2 τn−1 βn−1 τn−2 n+1 − βn−3 τn τn−3
aqn = − + . (72)
2 τn τn−2 2αn−2 τn−2 τn−1 τn
The above equations are compatible through (70). From the first equation in (9) we can get the
expression for the solution of the Painlevé XXIV equation (4) pn in terms of tau functions. Since
we have two different expressions for qn we have four different possibilities to write qn + qn+1 .
Of this four equations we write only the one being more compact, obtained by adding equation
(71) with (72) evaluated at n to n + 1. We get

en τn−1 τn+1 2m βn−1 τn−1 τn+1

pn = 2
=− . (73)
αn−1 τn a τn2

We can consider also the second equation in (9): this will give a fourth order quadratic equation
for the tau functions. Explicitly, from (9) and by using (73) and (72) we get
2 2
αn+1 τn+2

1 βn+2 τn+1 τn+4 βn τn τn+3 βn τn τn+2 βn+1 τn+1 τn+3
+ − 2 + + 2 + a2 (bz + c) = 0.
4 αn+1 τn+2 τn+3 τn+1 τn+3 αn+1 τn+1 τn+2 τn+1 τn+2
(74)
Notice that if we used equation (71) instead of (72) in the second equation (9) we would have
found the same quadratic equation for τn+4 in terms of τn+k , k = 0, ..., 3. Obviously, equations
(74) and (70) are compatible. Equation (74) can be also obtained directly by using (63) and one
of the two explicit equations for the Wronskian obtained above. The problem of the existence
of an equation of lower degree for the τ functions (but maybe of higher degree) seems to be a
difficult one to answer.
Finally, in this Section, we would like to present the explicit formula linking the Hamiltonian
function hn to the tau functions. From formula (2), by remembering that Hn = 2bmhn , and

11
from the above rational expressions of qn and pn in terms of tau functions it is possible to write
different equivalent formulae for hn . More explicitly, by using (73) and (71) we get:

2
hn 3 τ3
α2n−1 τn6 − 4βn−1 τn−1 n+1 a(bz + c)τn−1 τn+1 2
βn−2 τn−2 τn+1 2
− βn τn+2 τn−1 g2
= 4
− 2
− 3 4
− 2 .
en 8abmτn−1 τn τn+1 2bmαn−1 τn 8abmαn−1 τn−1 τn τn+1 4a ben
(75)

5 The elliptic case

If the two constants b and c are put equal to zero in equation (7), the equation degenerates in
the standard equation solved by the Weierstrass ζ function. The general solution of equation
(7) in this case is given by
 ae 2
. n
A + ζ(z + B, g2 , g3 ), g3 = − . (76)
2m
where ζ(z, g2 , g3 ) is the Weierstrass zeta function with invariants g2 and g3 , so that ζ̇ = −℘(z+B)
is an elliptic function of z. It is interesting to observe the dynamic of the Bäcklund transforma-
tions (20) on a seed solution of the general form (76). Notice that in this Section en+1 = en = e
since b = 0. Also, since (76) is the general solution, we expect that the action of the Bäcklund
transfoprmations is just a shift in z and in the dependent variable (i.e. a shift in the constants
A and B in (76)). Let us set

hn (z) = An + ζ(z + Bn , g2 , g3 ) − ζ(Bn , g2 , g3 ). (77)

The constant An fixes the value of hn at z = 0, whereas the derivative of hn (z) at z = 0 is given
by ḣn (0) = −℘(Bn , g2 , g3 ). The value of Bn is unique up to a shift in the corresponding period
parallelogram of the elliptic function and up to a sign. The map (20) implies the following
explicit map for An and implicit map for Bn :
℘(B
˙ n , g2 , g3 ) ae
An+1 = An + + (78)
2℘(Bn , g2 , g3 ) 4m℘(Bn , g2 , g3 )
ae℘(B
˙ n , g2 , g3 ) g2 g3
℘(Bn+1 , g2 , g3 ) = − − (79)
4m℘(Bn , g2 , g3 ) 4℘(Bn , g2 , g3 ) 2℘(Bn , g2 , g3 )2
By remembering that g3 = −a2 e2 /4m2 , the last expression can be rewritten as
ae 2
1 ℘(B
˙ n , g2 , g3 ) + 2m

℘(Bn+1 , g2 , g3 ) = − ℘(Bn , g2 , g3 ) (80)
4 ℘(Bn , g2 , g3 )
and recalls the addition formula for the Weierstrass ℘ function

˙ 2 , g2 , g3 ) 2
 
1 ℘(z
˙ 1 , g2 , g3 ) − ℘(z
℘(z1 + z2 , g2 , g3 ) = − ℘(z1 , g2 , g3 ) − ℘(z2 , g2 , g3 ) (81)
4 ℘(z1 , g2 , g3 ) − ℘(z2 , g2 , g3 )
By comparing equation (81) with (80) we can identify z1 with Bn and z2 with a zero of ℘(z, g2 , g3 ),
say z0 . Here we should remember that if z0 is a zero of ℘(z, g2 , g3 ) then −z0 is also a zero and
these are the only zeros in the given period parallelogram. From (7) with b = c = 0 one has
2 2
˙ 0 , g2 , g3 )2 = a4me2 . The choice of the sign of ℘(z
℘(z ˙ 0 , g2 , g3 ) amounts to the choice between direct
and inverse Bäcklund transformations. To fix the ideas, let us choose the sign in such a way
that the following equation is satisfied:
ae
℘(z
˙ 0 , g2 , g3 ) = − . (82)
2m

12
Then from (80) we have
 2
1 ℘(B
˙ n , g2 , g3 ) − ℘(z
˙ 0 , g2 , g3 )
℘(Bn+1 , g2 , g3 ) = − ℘(Bn , g2 , g3 ) = ℘(Bn + z0 , g2 , g3 ). (83)
4 ℘(Bn , g2 , g3 )

and we see that Bn+1 = Bn + z0 1 . This immediately gives

Bn = B0 + z0 n, ℘(z0 , g2 , g3 ) = 0. (84)

From the above discussion and equation (78) it follows that for n > 0 the values An are
explicitly given by
n−1
X ℘(B
˙ 0 + kz0 , g2 , g3 ) − ℘(z
˙ 0 , g2 , g3 )
An = A0 + . (85)
2℘(B0 + kz0 , g2 , g3 )
k=0

The sum in (85) can be explicitly performed by taking into account the addition theorems for
the Weierstrass ζ function, i.e. [17]

1 ℘(z,
˙ g2 , g3 ) − ℘(y,
˙ g2 , g3 )
= ζ(z + y, g2 , g3 ) − ζ(z, g2 , g3 ) − ζ(y, g2 , g3 ). (86)
2 ℘(z, g2 , g3 ) − ℘(y, g2 , g3 )

By using (86) we get:

An = A0 + ζ(Bn , g2 , g3 ) − ζ(B0 ) − nζ(z0 , g2 , g3 ). (87)

The τ function, related to hn by hn (z) = σ̇σnn , is proportional to the Weierstrass σ function,

related to the ζ function by σ̇ = ζσ. By choosing the normalization τn (0) = 1, explicitly we get

σ(z + Bn , g2 , g3 ) −z(ζ(Bn ,g2 ,g3)−An )

τn (z) = e (88)
σ(Bn , g2 , g3 )

The constants βn and αn defined in (55) and (56) can be easily found by using the classical
addition theorems of the Weierstrass σ function. We get

℘(z
˙ 0 , g2 , g3 )
αn = An+2 − An = − , βn = −℘(Bn+1 , g2 , g3 ). (89)
℘(Bn+1 , g2 , g3 )

From another point of view we can say that the classical addition theorems of the Weierstrass σ
function follow from the relation (55) and (56): these relations lift up the addition theorems of the
elliptic functions to solutions of Painlevé II equation. More will be said on this in Sections 8 and
(9). By taking into account ℘(z0 ) = 0 and ℘(z ˙ 0 ) = −ae/2m, the relation (57) also corresponds
to the addition formula for the Weierstrass function ℘. ˙ Also, by equating the relations (14) and
(73) for the solutions of the Painelvé XXXIV, i.e.
2m en τn−1 τn+1
pn = − ḣn , pn = (90)
a αn−1 τn2

one gets, in terms of elliptic functions, the following equality:

σ(z + Bn−1 , g2 , g3 )σ(z + Bn+1 , g2 , g3 ) σ(Bn−1 , g2 , g3 )σ(Bn+1 , g2 , g3 )

2
= = −σ(z0 , g2 , g3 )2 . (91)
σ(z + Bn , g2 , g3 ) ℘(z + Bn , g2 , g3 ) σ(Bn , g2 , g3 )2 ℘(Bn , g2 , g3 )
1
Obviously a choice has to be made on how to choose the signs: for example we can fix Re(Bn ) > 0 for all n
and Re(z0 ) > 0.

13
By setting z + Bn = y, the previous corresponds to the classical relation linking the σ functions
with the ℘ function [17]:

σ(y − z0 , g2 , g3 )σ(y + z0 , g2 , g3 )
= −℘(y, g2 , g3 ), ℘(z0 , g2 , g3 ) = 0. (92)
σ(y, g2 , g3 )2 σ(z0 , g2 , g3 )2

Finally we look at the relation (49). By setting z + Bn+1 = y we get

℘(y, g2 , g3 ) (ζ(y + z0 , g2 , g3 ) − ζ(y − z0 , g2 , g3 ) − 2ζ(z0 , g2 , g3 )) = −℘(z

˙ 0 , g2 , g3 ), (93)

i.e.
℘(z
˙ 0 , g2 , g3 )
Z
dy = ζ(y), (94)
ζ(y + z0 , g2 , g3 ) − ζ(y − z0 , g2 , g3 ) − 2ζ(z0 , g2 , g3 )
where obviously in the last two relations one must take into account that z0 is a zero of
℘(z, g2 , g3 ).

6 The rational case: recursions and relations for the Yablonskii-

Vorob’ev polynomials
It is well-known that there are special polynomials associated to solutions of the second Painelvé
equation. We will use the standard notation in the literature (see e.g. [1], [5], [14]). In this
Section we set the constants of the model as follows:
1
a = 1, b= , c = 0, en = n + 1/2, g2 = 0, m = 1. (95)
2
It is well known that for en = n + 1/2 it exists a unique rational solution of the Painlevé two
equation [16]. This solution can be obtained by the relations (12) and (53) as

τ̇n τ̇n+1
qn+1 = − . (96)
τn τn+1

The functions τn have the following structure in this case: the same exponential factor, given by
z3
e− 24 times a polynomial. The tau functions are defined up to an overall constant factor, so the
gauge must be fixed. In order to get the standard Yablonskii-Vorob’ev one has to set βn = −1/4
(and hence αn = 2n + 3 from equation (57)) in the Toda equation (56) (see also [14]). So, by
letting
. z3
τn (z) = Qn (z)e− 24 , (97)
from (56) we get the usual recurrence for the Yablonskii-Vorob’ev polynomials (see e.g. [1], [14]):
 
zQ2n+1 − 4 Qn+1 Q̈n+1 − Q̇2n
Qn+2 = , Q0 = 1, Q1 = z. (98)
Qn
Other differential-difference equations are:

Q̇n+2 Qn − Qn+2 Q̇n = (2n + 3)Q2n+1 (99)

and
Q̈n+1 Qn − 2Q̇n+1 Q̇n + Qn+1 Q̈n = 0. (100)

14
 Now all the equivalences found in Section (3) can be specialized by taking into account (95)
and (97). The Yablonskii-Vorob’ev polynomial satisfy the following fifth-order homogeneous
equation
16(n + 4)2 Q3n+3 16(n + 3)(n + 4)Qn+2 Qn+4 Q2n+4 Qn+1 (n + 4)Qn Q2n+3 Qn+4
Qn+5 = − − +
Q2n+2 Qn+1 (n + 3)Q2n+2 (n + 3)Qn+1 Q2n+2
(101)
and also the following fourth order recursion
2
Q2n+1 Qn+4 − Qn Q2n+3 (2n + 5)Q2n+2

Qn+1 Qn+3 Qn Qn+2
− = + − 2z. (102)
4(2n + 5)Qn+1 Qn+2 Qn+3 Qn+1 Qn+3 Q2n+2 Q2n+1

The rational solutions of the Painlevé II equation can be written in terms of the Yablonskii-
Vorob’ev polynomial as
(2n + 5)Q2n+2 Qn+4 Q2n+1 − Qn Q2n+3
qn+2 = − + , (103)
2Qn+1 Qn+3 8(2n + 5)Qn+1 Qn+2 Qn+3
or
(2n + 5)Q2n+2 Qn+4 Q2n+1 − Qn Q2n+3
qn+3 = − − (104)
2Qn+1 Qn+3 8(2n + 5)Qn+1 Qn+2 Qn+3
whereas the rational solutions of the Painlevé XXXIV equation are determined by
Qn Qn+2
pn+1 = . (105)
2Q2n+1

Also, they satisfy the following relation

ṗn+2 Qn+4 Qn+1 Qn+3 Qn
= − (106)
pn+2 4(2n + 5)Qn+2 Qn+3 4(2n + 5)Qn+1 Qn+2
Equations (101-105) have been given also in [2]. Here we have shown how these relations are
not peculiar of the Yablonskii-Vorob’ev polynomial but come from the more general structure
underlying solutions of the Hamiltonian equation (7) described in Section (3).

7 Airy-type solutions
As we have seen at the beginning of Section (3), it is possible to get solutions of the Painlevé
equation II in terms of Airy functions and their derivatives if en = 2abmn, n ∈ N. Let us make
a more precise statement. In general the parameter en is defined to be equal to e + 2abmn, with
e ∈ R or C. In this Section we set e = 0. Also, we assume b 6= 0. Similar to the rational case,
g2
the tau functions have the following structure: the same exponential factor, given by e− 4a2 b z
times a polynomial of Airy functions and their derivatives. We remember equation (8) and set:
g2 g2
τn (z) = e− 4a2 b z Ψn (z), hn (z) = − + kn (z), (107)
4a2 b
where the relation between Ψn and kn is

Ψ̇n
kn = . (108)
Ψ
It is easily see that the first two Airy polynomials are given by
.
Ψ0 (z) = 1, Ψ1 (z) = Ψ̈1 + a2 (bz + c)Ψ1 = 0. (109)

15
  2 1/3

The function Ψ1 can be explicitly written as a linear combination of Ai − (a b) b (bz+c) and
 2 1/3

Bi − (a b) b (bz+c) , where Ai(z) is, up to a constant factor, the only solution of the Airy differ-
ential equation ÿ = zy exponentially decaying on the positive real axis. The function Bi(z) is
instead exponentially increasing on the positive real axis (see e.g. [19, 20]). In order to get stan-
dard sequence of Airy-type solutions one shloud set βn = −1 in equation (56) (see for example
[1]). The corresponding recursion for Ψn is

−Ψn+1 Ψ̈n+1 + Ψ̇2n+1

Ψn+2 = . (110)
Ψn
Since βn is fixed, the value of αn is fixed from (57) and we get αn = a2 b(n + 1), so that the
relation (55) can be rewritten as

Ψ̇n+2 Ψn − Ψ̇n Ψn+2 = a2 b(n + 1)Ψ2n+1 . (111)

The function k1 solves a Riccati equation, i.e. a first order equation. Indeed we expect that
all the functions kn belonging to the Airy-type solutions of (7) solve a first order differential
equation. By starting from the equation for k1 , i.e.

k̇1 + k12 + a2 (bz + c) = 0, (112)

it is possible to show, by using the Bäcklund transformations and equation (112) itself that k2
solves the following differential equation:
 2
k̇2 + k22 + 4a2 k2 ((bz + c)k2 − b) = 0. (113)

For k3 one obtains a first order equation cubic in k̇3 . There are other differential-difference
equations solved by the Airy-type solutions. Indeed we have
 
a2 b (n + 1)Ψ̇n Ψn+1 − nΨ̇n+1 Ψn + Ψn+2 Ψn−1 = 0 (114)

and
Ψ̈n+1 Ψn − 2Ψ̇n Ψ̇n+1 + Ψ̈n Ψn+1 + a2 (bz + c)Ψn+1 Ψn = 0. (115)
The solution of the Painleve II equations (4) are related to the functions Ψ by
Ψn+4 Ψ2n+1 − Ψn Ψ2n+3 a2 b(n + 2)Ψ2n+2
2aqn+2 = − , (116)
a2 b(n + 2)Ψn+1 Ψn+2 Ψn+3 Ψn+1 Ψn+3
or
Ψn+4 Ψ2n+1 − Ψn Ψ2n+3 a2 b(n + 2)Ψ2n+2
2aqn+3 = − − , (117)
a2 b(n + 2)Ψn+1 Ψn+2 Ψn+3 Ψn+1 Ψn+3
whereas the solutions of the Painlevé XXXIV equation are given by
2m Ψn−1 Ψn+1
pn = (118)
a Ψ2n
and satisfies
ṗn Ψn+2 Ψ2n−1 − Ψn−2 Ψ2n+1
= . (119)
pn a2 bnΨn−1 Ψn Ψn+1
The fifth order explicit recursion (70) now gives
a4 b2 (n + 3)2 Ψ3n+3 Ψ2n+4 Ψn+1 a4 b2 (n + 3)(n + 2)Ψn+4 Ψn+2 (n + 3)Ψn+4 Ψ2n+3 Ψn
Ψn+5 = − − +
Ψ2n+2 (n + 2)Ψ2n+2 Ψn+1 (n + 2)Ψ2n+2 Ψn+1
(120)

16
8 The general case
In this Section we consider the full equation (7) with b 6= 0 and assume that en is different from
the values considered in Sections (6) and (7). Let us investigate the solutions of (7) that are not
singular in z = 0. As we will show, it is possible to parametrize the coefficients of the Taylor
series of solutions of equation (7) in terms of Weierstrass elliptic functions. By generalizing the
construction given in Section (5), we will look at the evolution of the coefficients in the Taylor
series under the action of the Bäcklund map. As we did in Section (5), let us set:

hn (z) = An + ξ(z + Bn ) − ξ(Bn ), (121)

where An is equal to hn (0) and ḣn (0) = ξ̇(Bn ). The series around z = 0 reads
X Cn,k dk ξ(z) .
hn (z) = zk , Cn,k = , k = 1, 2, . . . Cn,0 = An (122)
k! dz k z=Bn
k=0

and, by taking the derivative of equation (7) and looking at the corresponding third order
equation for hn (z) one has that the coefficients Cn,k satisfy the following quadratic recursion for
m ≥ 0:
m+1
X m g2
Cn,m+3 = −6 Cn,k+1 Cn,m+1−k − 2a2 ((2m − 1)bCn,m + 2cCn,m+1 ) + δ0,m . (123)
k 2
k=0

Only two of the first three coefficients are independent since they must be related by the following
equation:
2 3 a2 e2n
Cn,2 + 4Cn,1 + 4a2 c Cn,1
2
− (4a2 b An + g2 )Cn,1 − =0 (124)
4m2
The previous can be seen as an elliptic curve and can be uniformized in terms of elliptic functions.
More specifically we set:
a2 c p p p p
Cn,1 = − − ℘(Bn , gn,2 , gn,3 ), Cn,2 = −℘(B
˙ n , gn,2 , gn,3 ) (125)
3
p p
where gn,2 and gn,3 are the deformed Painelvé invariants defined by

p a2 c2 p aen 2 8a6 c3 4a4 bcAn a2 cg2

gn,2 = g2 + 4a2 (bAn + ), gn,3 = −( ) + + + . (126)
3 2m 27 3 3
It is convenient to identify the constant c with a Weierstrass ℘ function too. Let us introduce
the point tn such that
3 p p
c = − 2 ℘(tn , gn,2 , gn,3 ). (127)
a
The point tn is defined up to a sign within a period parallelogram. From equations (126) we get
 ae 2
p p 2 n
℘(t
˙ n , gn,2 , gn,3 ) = , (128)
2m
and, as we did in Section (5), we fix the sign in such a way that
p p aen
℘(t
˙ n , gn,2 , gn,3 )=− . (129)
2m
With this choice, the Bäcklund transformations (20) define the following recursions for the
constants An and Bn :
p p p p
1 ℘(B
˙ n , gn,2 , gn,3 ) − ℘(t
˙ n , gn,2 , gn,3 )
An+1 = An + p p p p , (130)
2 ℘(Bn , gn,2 , gn,3 ) − ℘(tn , gn,2 , gn,3 )

17
and
p p p p
℘(Bn+1 , gn+1,2 , gn+1,3 ) = ℘(Bn + tn , gn,2 , gn,3 ). (131)
The explicit value of An from the recursion (130) can be simplified by using (86) and (131). One
gets
n−1
p p p p p p
X
An = A0 + ζ(Bn , gn,2 , gn,3 ) − ζ(B0 , gn,2 , gn,3 ) − ζ(tk , gk,2 , gk,3 ). (132)
k=0

In order to write explicitly the recursion (70) and the expressions (71), (72) and (73) relating
the solutions of the Painelvé II equation and Painlevé XXXIV equation to the tau functions it
is important to have the values of the constants αn and βn defined by equations (55) and (56).
By fixing the gauge τn (z) = 1 + O(z), by using the explicit recursion (123) and the definition
(53) of the tau functions we can calculate:
p p p p
αn = An+2 − An , βn = ℘(tn+1 , gn+1,2 , gn+1,3 ) − ℘(Bn+1 , gn+1,2 , gn+1,3 ) (133)

All the previous expressions reduce to those of Section (5) for b = c = 0. We notice that for
b = 0 and c 6= 0 equation (7) is still explicitly integrable in terms of elliptic functions: the
invariants (126) do not depend on n like those of Section (5) in this case, the point tn also is
fixed, i.e. independent on n and the relation (131) for Bn is explicitly solved in Bn = B0 + tn.
As a final remark, we notice that until now we investigated the action of Bäcklund trans-
formations on solutions of (7) regular at the origin z = 0. The dynamics of the Bäcklund
transformations on solutions having a singular point in z = 0 is interesting and deserve a sep-
arate work. Let us only notice that the singular points, i.e. the zeros of τn , correspond to the
zeros of ḣn+1 : the singularity develop in a regular solution with the coefficients of z in the Taylor
series being equal to zero. This is clear also by writing equations (55) and (49) as:
2
τn+1
hn+2 − hn = αn (134)
τn τn+2

and
τn τn+2
ḣn+1 = βn 2 . (135)
τn+1
Equations (134) and (135) show how the zeros and poles of the Hamiltonian functions hn ’s are
interlaced: these aspects also will be taken into account in a separate work.

9 Integral relations
In this Section we would like to present certain integral relations satisfied by the tau functions
τn and the Hamiltonian functions hn .
Let us look again at the Bäcklund transformations (20) and (23). Let us define the functions
Fn (z) as the integral of the following function

aen τn2 τn2

Z Z
Fn (z) = dz = −αn−1 dz, (136)
2m τ̈n τn − τ̇n2 τn−1 τn+1

where the path of integration avoids the zeros of τ̈n τn − τ̇n2 (i.e. those of τn−1 τn+1 ). By integrating
(20) we get
2
τn+1
= e−Fn (137)
τ̈n τn − τ̇n2

18
and from (56)
τn+1 = βn−1 τn−1 e−Fn (138)
The inverse transformation gives the same result, i.e. τn−1 proportional to τn+1 eFn . Indeed,
from (23) one has
τn2
2 = eFn+1 = eFn+1 , (139)
τ̈n+1 τn+1 − τ̇n+1
i.e.
τn = βn τn+2 eFn+1 . (140)
The constant on the right hand side of (140) and (138) depends on the path of integration chosen
and on the fixed gauge for the τn functions. In general we could write

τn+1 = cn τn−1 e−Fn (141)

for some constant cn . Equation (141) expressing the integral (136) in terms of tau functions can
be seen as the equivalent of the Bäcklund transformation (20) for the tau function τn . Other
integral relations can be obtained by the results given in Section (8). For example from (55)
one has: Z 2
τn+2 τn+1
= αn dz + Cn , (142)
τn τn2
for some constant Cn . Also, the Hamiltonian function itself can be expressed through two
integral relations involving other Hamiltonian functions or the tau functions. Indeed one has
αn−1 βn−1 βn−1 τn−1 τn+1
Z Z
hn = dz = dz. (143)
hn+1 − hn−1 τn2

The relations given in this Section are general: the first element of the chain h0 , or τ0 can
be a rational solution, a special type solution or a transcendental solution. The successive
Hamiltonians can be obtained by means of (20), whereas the successive tau functions for example
by means of (56).

10 Conclusions
The generalized structures presented in this work suggest that the solutions of the Painlevé II
equation, Painlevé XXXIV equation, the Yablonskii-Vorobev polynomials, the Airy type solu-
tions of the Painlevé II, the Weierstrass elliptic functions and the Hamiltonians of the Painlevé
II equation can be considered the members of an extended Painlevé II hierarchy. There are dif-
ferent points not yet discussed in this paper and that will be the subject of future works: from
the interlacing of the poles and zeros of the Hamiltonians, to some properties of the Yablonskii-
Vorobev that follow for example from equation (99). Finally, we believe that the structures here
described could be lifted-up at the Painlevé equation IV: this also will be the subject of future
research.

Acknowledgments

I wish to acknowledge the support of Università degli Studi di Brescia; INFN, Gr. IV - Mathe-
matical Methods in NonLinear Physics and ISNMP - International Society of Nonlinear Math-
ematical Physics.

19
References
[1] P.A. Clarkson: Painlevé Equations - Nonlinear Special Functions. In: Marcellán, F., Van Assche,
W. (eds) Orthogonal Polynomials and Special Functions. Lecture Notes in Mathematics, vol 1883.
Springer, Berlin, Heidelberg. (2006)

[2] P. A. Clarkson: Remarks on the Yablonskii-Vorob’ev polynomials Physics Letters A 319 (2003)
137-144.

[3] P.A. Clarkson: Open Problems for Painlevé Equations, SIGMA, Symmetry, Integrability and Ge-
ometry: Methods and Applications, 15 (2019), 006, 20 pages

[4] P.J. Forrester, N.S. Witte: Application of the τ -Function Theory of Painlevé Equations to Random
Matrices: PIV, PII and the GUE, Commun. Math. Phys. 219 (2001) 357.

[5] S. Fukutani, K. Okamoto and H. Umemura, Special polynomials and the Hirota bilinear relations
of the second and fourth Painlevé equations, Nagoya Math. J., 159 (2000) 179-200

[6] : V. I. Gromak, I. Laine and S. Shimomura: Painlevé Differential Equations in the Complex Plane,
de Gruyter Studies in Mathematics, Walter de Gruyter GmbH & Co., Berlin, Germany.

[7] A.N.W. Hone and F. Zullo: Properties of the series solution for Painlevé I, J. Nonlinear Math. Phys.
20 (Suppl. 1), 85-100, (2013)

[8] A.N.W. Hone and F. Zullo: A Hirota bilinear equation for Painlevé transcendents PIV , PII and PI ,
Random Matrices: Theory and Applications, Vol. 07, No. 04, 1840001 (2018)

[9] E. L. Ince, Ordinary Differential Equations, Dover, New York (1956).

[10] K. Okamoto: On the τ -functions of the Painlevé equations, Physica 2D, 525-535, (1981)

[11] K. Okamoto: Studies on the Painlevé equations, Math. Ann. 275, 221-255 (1986)

[12] A.P. Vorob’ev, On rational solutions of the second Painlevé equation, Diff. Eqns., 1 (1965) 58-59

[13] A.I. Yablonskii, On rational solutions of the second Painlevé equation, Vesti Akad. Navuk. BSSR
Ser. Fiz. Tkh. Nauk., 3 (1959) 30-35

[14] M. Taneda, Remarks on the Yablonskii-Vorob’ev polynomials, Nagoya Math. J., 159 (2000) 87-111

[15] H. Umemura, Painlevé equations and classical functions, Sugaku Expositions, 11 (1998) 77-100

[16] H. Umemura and H. Watanabe, Solutions of the second and fourth Painlevé equations I, Nagoya
Math. J., 148 (1997) 151-198.

[17] E. T. Whittaker, G.N. Watson: A course of modern analysis, Cambridge University Press, Cam-
bridge, 1927.

[18] F. Zullo: Bácklund transformations and Hamiltonian flows, J. Phys. A: Math. Theor. 46 145203,
2013

[19] F. Zullo: On the solutions of the Airy equation and their zeros, Complex Differential and Difference
Equations: Proceedings of the School and Conference held at Bedlewo, Poland; G. Filipuk, A.
Lastra, S. Michalik, Y. Takei and H. Zoladek editors, 2020

[20] F: Zullo: On the dynamics of the zeros of solutions of the Airy equation, Mathematics and Computers
in Simulation (MATCOM), Elsevier, vol. 176(C), pages 312-318

20