GEOMETRIC PROGRESSIONS IN THE SETS OF VALUES OF

RATIONAL FUNCTIONS

MACIEJ ULAS
arXiv:2304.09264v1 [math.NT] 18 Apr 2023

Abstract. Let a, Q ∈ Q be given and consider the set G(a, Q) = {aQi : i ∈

N} of terms of geometric progression with 0th term equal to a and the quotient
Q. Let f ∈ Q(x, y) and Vf be the set of finite values of f . We consider the
problem of existence of a, Q ∈ Q such that G(a, Q) ⊂ Vf . In the first part of the
paper we describe several classes of rational function for which our problem has
f (x,y)
a positive solution. In particular, if f (x, y) = f1 (x,y) , where f1 , f2 ∈ Z[x, y]
2
are homogenous forms of degrees d1 , d2 and |d1 − d2 | = 1, we prove that
G(a, Q) ⊂ Vf if and only if there are u, v ∈ Q such that a = f (u, v). In the
second, experimental, part of the paper we study the stated problem for the
rational function f (x, y) = (y 2 −x3 )/x. We relate the problem to the existence
of rational points on certain elliptic curves and present interesting numerical
observations which allow us to state several questions and conjectures.

Dedicated to Jaap Top on the occasion of his 60 + 2 + ε birthday.

1. Introduction and motivation

The problem of finding all integer solutions of the Diophantine equation
x2 + 7 = 2n
is a classical one. In fact, the above equation is the famous Ramanujan-Nagell
equation. Ramanujan conjectured that the equation has only five solutions corre-
sponding to n = 3, 4, 5, 7, 15 [8]. This was proved by Nagell in [6]. Today, by a
Ramanujan-Nagell equation we understand a general equation of the form
x2 + D = aQn ,
where a, Q, D ∈ Z are fixed. The theory of such equations is quite well developed
and we have many explicit results concerning form of solutions of particular equa-
tions or the number of integer solutions (see for example [2, 3] and references given
therein). It is clear that one can consider even more general equation of the form
(1) f (x) = aQn ,
where f is a given polynomial with rational coefficients and a, Q ∈ Z are fixed. We
know that if f has at least two roots in Q, then (1) has only finitely many solutions
in integers x, n. In fact, even if we allow Q to vary, then, under mild conditions
on f , the equation (1) has only finitely many solutions in integers x, Q, n. More on
this line of research, or to be more precise, the realm of this type of Diophantine
equations, can be found in the monograph [9].

2000 Mathematics Subject Classification. 11D41.

Key words and phrases. rational function, geometric progression, elliptic curve.
The research of the authors is supported by the grant of the National Science Centre (NCN),
Poland, no. UMO-2019/34/E/ST1/00094.
1
2 MACIEJ ULAS

Let us observe that the result concerning the solvability of (1), where a, Q are
fixed, can be phrased in a different way. More precisely, let us consider the set
G(a, Q) = {aQi : i ∈ N},
where a, Q ∈ Q, aQ(Q2 − 1) 6= 0, i.e., an infinite set of values of geometric pro-
gression with initial term equal to a and the quotient equal to Q, and the set
Wf = {f (x) : x ∈ Z} - the set of values of the polynomial f . In other words, the
equation (1) is solvable in integers if and only if G(a, Q) ∩ Wf is non-empty. From
our discussion it follows that if f ∈ Q[x], then the only possibility for G(a, Q) ∩ Wf
to be infinite is when f is a power of a linear polynomial. Moreover, if f is linear
polynomial then it is not difficult to construct infinitely many pairs a, Q ∈ Z such
that the stronger condition G(a, Q) ⊂ Wf holds. Indeed, let f (x) = Ax + B, where
A, B are fixed and observe that for each t, T ∈ Z we have G(f (t), AT + 1) ⊂ Wf .
This is a consequence of the identity f (vn ) = f (t)(AT + 1)n , where v0 = f (t) and
for n ≥ 1 we have vn = (AT + 1)vn−1 + BT .
Let us change a perspective and consider polynomials with more variables. Then,
it is possible that G(a, Q) is contained in the set of values of f even if the degree
of f is ≥ 2. For example, if p ≡ 1 (mod 4) is a prime number, then the set
G(1, p) = {pn : n ∈ N} is contained in the set of values of the polynomial x2 + y 2 .
This simple observation motivated us to consider, for a given f ∈ Q(x, y), the set
Vf = {f (u, v) ∈ Q : u, v ∈ Q for which f (u, v) determined}
and study the following general problem.
Problem 1.1. Let f ∈ Q(x, y) be given. Does there exist rational numbers a, Q
such that G(a, Q) ⊂ Vf ?
Let us note that the above problem in certain cases can be seen as a variant
of a well studied problem. More precisely, if f ∈ Z[x, y] is a homogenous form
of degree d ≥ 3, then for given a, Q ∈ Z the equation f (x, y) = aQi is a special
case of a so called Thue-Mahler equation. It particular, it has only finitely many
solutions (x, y, i) in co-prime integers x, y and i ∈ N. This was proved by Mahler in
[5] (generalizing an earlier result of Thue [10] concerning the equation f (x, y) = A,
where A ∈ Z \ {0} is fixed). However, we do not assume that x, y are integers.
This assumption makes the problem more interesting and, in certain cases, allows
to use the techniques typically applied in Diophantine problems where there is no
real restriction between rational and integral solutions (like in case of homogenous
problems).
Let us describe the content of the paper in some details. In Section 2 we present
several classes of rational functions for which Problem 1.1 has a solution. We also
introduce a distinction between solutions of Problem 1.1 in the case of weighted ho-
mogenous functions of degree d. In Section 3 we give some observations concerning
Problem 1.1 for the rational function f (x, y) = (y 2 − x3 )/x and shows connections
with the existence of rational points on certain elliptic curve (Theorem 3.1). As a
consequence of our investigations, we prove that for each a ∈ {1, . . . , 103 } such that
the elliptic curve y 2 = x3 + ax has positive rank, there are infinitely many values of
Q ∈ Q such that Q is not a square and G(a, Q) ⊂ Vf . Finally, in the last section we
consider Problem 1.1 for the rational function f (x, y) = (y 2 − x3 )/x from a com-
putational point of view. We present results of various computer experiments and
 GEOMETRIC PROGRESSIONS IN THE SETS OF VALUES OF RATIONAL FUNCTIONS 3

formulate several questions, problems and conjectures which may stimulate further
research .
During our computational experiments we extensively used two computational
packages: the latest version of PARI/GP [7] (in case of rank computations) [7] and
Mathematica 13.2 (for symbolic computations and data manipulation) [13].

2. Basic results
In this section we present certain classes of rational functions such that for each
element f of an appropriate class there are values a, Q ∈ Q such that G(a, Q) ⊂ Vf .
Theorem 2.1. Let g1 , g2 , h1 , h2 ∈ Q(y) and suppose that g1 h2 6= h1 g2 in Q(y). Let
xg1 (y) + g2 (y)
f (x, y) = .
xh1 (y) + h2 (y)
Then, for any a ∈ Q \ {0} and Q 6= ±1 we have G(a, Q) ⊂ Vf .
Proof. Let a, Q be given. For given i ∈ N we need to find rational numbers xi , yi
such that f (xi , yi ) = aQi for i = 0, 1, . . .. From our assumptions, for any i ∈ N, we
can chose yi ∈ Q such that aQi h1 (yi ) − g1 (yi ) 6= 0. Then, it is enough to take
g2 (yi ) − aQi h2 (yi )
xi =
aQi h1 (yi ) − g1 (yi )
and note that f (xi , yi ) = aQi for i ∈ N. 
Theorem 2.2. Let f (x, y) = ux2 + vxy + wy 2 ∈ Q[x, y] \ {0}. There are infinitely
many pairs a, Q ∈ Q such that G(a, Q) ⊂ Vf is infinite.
Proof. Multiplying f by 4u we get that 4uf (x, y) = (2ux + vy)2 + (4uw − v 2 )y 2 .
We substitute a = 4ua0 and write X = 2ux + vy, Y = y, d = 4uw − v 2 . Thus, the
problem of finding a, Q ∈ Q such that G(a, Q) ⊂ Vf is equivalent with the problem
of finding a0 , Q ∈ Q such that G(a0 , Q) ⊂ VF , where F (x, y) = x2 + dy 2 . To get the
result we take a0 = r2 + ds2 , Q = u2 + dv 2 , where r, s, u, v ∈ Q are chosen in such
a way that rsuvF (r, s)F (u, v) 6= 0. Then, noting that the form F is multiplicative,
we get the result. 
Theorem 2.3. Let f (x, y) = ff21 (x,y)
(x,y) , where f1 , f2 ∈ Z[x, y] are homogenous forms
of degrees d1 , d2 , respectively. Let us assume that |d1 − d2 | = 1. Then, for any
Q ∈ Q \ {−1, 0, 1} we have G(a, Q) ⊂ Vf if and only a = f (u, v) for some u, v ∈ Q.
Proof. Without loss of generality we can assume that d1 − d2 = 1. If G(a, Q) ⊂ Vf ,
then for some u, v ∈ Q we have a = f (u, v). To get the implication in the other
side, we take xi = uQi , yi = vQi , where u, v are rational parameters. Then we have
f1 (uQi , vQi ) Qd1 i f1 (u, v)
f (xi , yi ) = = = Qi f (u, v).
f2 (uQi , vQi ) Qd2 i f2 (u, v)
Thus, if a = f (u, v) for some u, v ∈ Q such that f (u, v) 6= 0 we have f (xi , yi ) = aQi
for each i ∈ N. Consequently G(a, Q) ⊂ Vf . 
Remark 2.4. Theorem 2.3 is a special (easily) provable part of the following
general observation. Suppose that f ∈ Q(x, y) is weighted homogeneous of degree
d, i.e., there is a pair (w1 , w2 ) of integers, called weights, with such a property that
for a variable λ, we have f (λw1 x, λw2 y) = λd f (x, y). Thus, to find a, Q ∈ Q such
4 MACIEJ ULAS

that G(a, Q) ⊂ Vf it is enough to prove that we have {a, aQ, . . . , aQd−1 } ⊂ Vf . This
is clear, if f (xi , yi ) = aQi for i = 0, 1, . . . , d − 1, then f (xi Qw1 j , yi Qw2 j ) = aQi+dj
for j ∈ Z and thus G(a, Q) ⊂ Vf .
Note that a rational function f is homogenous of degree d if and only if is
weighted homogenous of degree d with weights (w1 , w2 ) = (1, 1).
Example 2.5. Let f (x, y) = x3 + y 3 . We find that G(13797, 2) ⊂ Vf which follows
from observation made in Remark 2.4 and the equalities
f (−3, 24) = 13797, f (−13, 31) = 2 · 13797, f (−33, 45) = 22 · 13797.
Let us also note that the problem of finding a, Q such that G(a, Q) ⊂ Vf can be
related to the Problem 1.1 with different rational function f . Indeed, let us recall
that the curve C : x3 + y 3 = A is birationally equivalent with the elliptic curve in
Weierstrass form E : y 2 = x3 − 432A2 . We have the maps
 
36A + y 36A − y
ϕ : C 3 (x, y) 7→ , ∈ E,
6x 6x
 
12A 36A(y − x)
ψ : E 3 (x, y) 7→ , ∈ C.
x+y x+y
In other words, if g(x, y) = y 2 − x3 , then we have
G(a, Q) ⊂ Vf ⇐⇒ G(−432a2 , Q2 ) ⊂ Vg .
Let us consider a = 6, Q = 7, i.e., for i = 0, 1, 2 we consider the curve Ei : y 2 =
x − 432 · 62 · 72i . One can check that the point Si lies on the curve Ei for i = 0, 1, 2,
3

where
S0 = (28, 80), S1 = (172, 2080), S2 = (2353, 113975).
Then, we have x3i + yi3 = 6 · 7i for the corresponding points Pi = ψ(Si ) = (xi , yi ),
where
     
37 17 449 71 124559 103391
P0 = , , P1 = ,− , P2 = ,− .
21 21 129 129 14118 14118
As a consequence we get the following
Corollary 2.6. Let C = {x3 + y 3 : x, y ∈ Z} ⊂ Z. The set C contains arbitrarily
long geometric progressions.
Proof. Let n ∈ N+ be given and f (x, y) = x3 +y 3 . We proved that the set G(6, 7) ⊂
i
Vf . Let xi , yi ∈ Q \ {0}
Qn satisfy f (xi , yi ) = 6 · 7 for i ∈ N. Let Dn be a non-zero
integer such that Dn i=0 (xi yi ) ∈ Z. Then the numbers Xi = Dn xi , Yi = Dn yi
are integers and we have
f (Xi , Yi ) = 6Dn3 7i for i = 0, 1, . . . , n,
i.e., the set C contains geometric progression {6Dn3 7i : i = 0, . . . , n} and the result
follows. 

In the light of the proof of Corollary 2.6 on can ask the following
Question 2.7. Do there exist integers a, Q such that for each i ∈ N there is a
co-prime solution in integers x, y of the Diophantine equation x3 + y 3 = aQi ?
 GEOMETRIC PROGRESSIONS IN THE SETS OF VALUES OF RATIONAL FUNCTIONS 5

We should also note that in the case of f ∈ Q(x, y) which is weighted homogenous
of degree d with weights (w1 , w2 ), the Problem 1.1 has a trivial solution. Indeed,
for each Q ∈ Q, Q(Q2 − 1) 6= 0, and any u, v ∈ Q such that f (u, v) 6= 0, we have
G(f (u, v), Qd ) ⊂ Vf . This is simple consequence of the identity
f (uQiw1 , vQiw2 ) = f (u, v)Qid .
Thus, in case of weighted homogenous rational function f of degree d, we need to
consider various notions of solutions of Problem 1.1.
Definition 2.8. Let f ∈ Q(x, y) be weighted homogenous of degree d. We say that
Problem 1.1 has
(1) a trivial solution (a, Q) if and only if Q is a d-th power of a rational number
and G(a, Q) ⊂ Vf ;
(2) a non-trivial solution (a, Q) of level e if and only if e|d, e < d, Q is e-th
power of a rational number, there is at least one prime q such that the
q-adic valuation of Q is equal to e (or −e), and G(a, Q) ⊂ Vf ;
(3) a proper solution (a, Q) if and only if (a, Q) is a non-trivial solution of level
1.
We already observed that for each weighted homogenous rational function f ,
Problem 1.1 has a trivial solution. On the other hand, for each f from the class of
homogenous rational functions of degree 1, we know, via Theorem 2.3, that Problem
1.1 has a proper solution. The same is true for the function f (x, y) = x3 + y 3 . In
the next section we will investigate the existence and properties of proper solutions
of Problem 1.1 for f (x, y) = (y 2 − x3 )/x. However, before we do that we note the
following.
Theorem 2.9. Let d ∈ N be odd.
(1) Let f (x, y) = (y 2 − xd )/x. For each Q ∈ Q \ {−1} which is not a square,
there are infinitely many values of a ∈ Z such that (a, Qd−1 ) is a non-trivial
solution of level d − 1 of Problem 1.1.
(2) Let f (x, y) = y 2 − xd . For each Q ∈ Q \ {−1} which is not a square,
there are infinitely many values of a ∈ Z such that (a, Qd ) is a non-trivial
solution of level d of Problem 1.1.
Proof. We start with the case f (x, y) = (y 2 − xd )/x. In this case f is weighted
homogenous of degree 2(d − 1) with weights (2, d). Let us note that to get the
statement G(a, Qd−1 ) ⊂ Vf it is enough to prove that there exist rational solutions
of the system of equations
f (x0 , y0 ) = a, f (x1 , y1 ) = aQd−1
in variables a, xi , yi , i = 0, 1. Equivalently, by eliminating a, we consider the equa-
tion
f (x1 , y1 ) = f (x0 , y0 )Qd−1 .
It can be easily solved via the substitution xi = pi T, yi = qi T m , where m =
(d − 1)/2, pi , qi for i = 0, 1 are rational variables and we look for an appropriate
specialization of T . Indeed, after the substitution, our equation reduces to the
equation (in one variable T ) of the form
T d [p0 p1 (pd−1
1 − pd−1
0 Qd−1 )T + p1 q02 Qd−1 − p0 q12 ] = 0.
6 MACIEJ ULAS

Solving the equation defined by the second factor, we get an appropriate value of
T , together with the corresponding value of a in the following form
p0 q12 − p1 q02 Qd−1 T d−2 (q02 − pd0 T )
T = , a = f (x0 , y0 ) = f (p0 T, q0 T m ) = .
p0 p1 (pd−1
1 − pd−1
0 Qd−1 ) p0
i
As a consequence of our reasoning we see that by taking Xi = xi (mod 2) Q2b 2 c , Yi =
i
xi (mod 2) Qdb 2 c we get f (Xi , Yi ) = aQ(d−1)i for i ∈ N, and thus G(a, Qd−1 ) ⊂ Vf .
If Q is not a square then our solution is non-trivial.

To get the statement for f (x, y) = y 2 −xd we note that f is weighted homogenous
of degree 2d with weights (2, d). It is clear that to prove that G(a, Qd ) ⊂ Vf it is
enough to solve the system
f (x0 , y0 ) = a, f (x1 , y1 ) = aQd .
or equivalently, the equation f (x1 , y1 ) = f (x0 , y0 )Qd . Using the same substitutions
and the same reasoning as in the case of the function (y 2 − xd )/x we get the
corresponding values of T and a in the following form
q02 − Qd q12
T = , a = f (x0 , y0 ) = f (p0 T, q0 T m ) = T d−1 (q02 − pd0 T ).
pd0 − Qd pd1

y 2 −x3
3. Geometric progressions in the set Vf for f (x, y) = x and
related elliptic curves
In the previous section we obtained some general result concerning Problem 1.1.
Now we concentrate on a specific rational function. Let us put
y 2 − x3
f (x, y) = .
x
We know that f is weighted-homogenous of degree 4 with weights (2, 3) and from
the first part of Theorem 2.9, for each Q ∈ Q which is not a square we have
G(a, Q2 ) ⊂ Vf non-trivially. Before we will go on, let us note that without loss of
generality one can assume that a, Q ∈ Z. Indeed, if a = p/q, Q = u/v, then by
multiplication of the equality y 2 = x3 + aQi x by q 6 v 6i we easily get the equivalence
G(a, Q) ⊂ Vf ⇐⇒ G(pq 3 , uv 3 ) ⊂ Vf .
Moreover, let us note that for any t ∈ Q we have: (a, Q) is a solution of Problem
1.1 if and only if (at4 , Q) is a solution of Problem 1.1.
Although the choice of f may be seen arbitrary, it is not. This follows from the
fact that for any given A 6= 0 the curve f (x, y) = A is an elliptic curve, and the
existence of a, Q such that G(a, Q) ⊂ Vf is equivalent with the property that for
each i ∈ N, the elliptic curve
Ei (a, Q) : y 2 = x3 + aQi x
contains rational point with x 6= 0. Actually, in the sequel we require that the curve
has a positive rank. This connection with elliptic curves suggest several interesting
questions.
It is not difficult to find an example of rational numbers a, Q such that G(a, Q) is
infinite and G(a, Q) ⊂ Vf . Indeed, let us take Q = 2 and a = 47. We thus consider
elliptic curves Ei (47, 2) : y 2 = x3 + 47 · 2i x. One can compute the rank of Ei (47, 2)
 GEOMETRIC PROGRESSIONS IN THE SETS OF VALUES OF RATIONAL FUNCTIONS 7

and the set of generators of infinite order. We collect the computed data in the
table below.
i r(Ei (47, 2)) generators of Ei (47, 2)(Q)
0 1 (289/25, −5712/125)
1 2 (2, 14), (1504/81, 65800/729)
2 1 (18, 96)
3 1 (4716544/18225, 10271916928/2460375)
Table 1. Ranks and the generators for the elliptic curves
Ei (47, 2) : y 2 = x3 + 47 · 2i x, i = 0, 1, 2, 3.
As a consequence we get that G(47, 2) ⊂ Vf . In the light of this result one can
consider related problem. More precisely, let a ∈ Q \ {0} be given and assume that
the elliptic curve E : y 2 = x3 + ax has a positive rank. Is it possible to find a
Q ∈ Z \ {−1, 0, 1} which is fourth power-free and the elliptic curve Ei (a, Q) has
positive rank for i = 0, 1, 2, 3? We prove the following:
Theorem 3.1. Let a ∈ Z \ {0} and suppose that the elliptic curve E : y 2 = x3 + ax
has a positive rank. Let us consider the elliptic curve
Ea (p, v) : Y 2 = X(X − f1 (a, p, v))(X − f2 (a, p, v)) =: F (X),
where
f1 = ap3 (p2 + a)3 v 4 ,
f2 = ap(p2 + a)((p2 + a)pv 2 − 1)((p2 + a)pv 2 + 1).
If there are p, v ∈ Q such that Ea (p, v) has a positive rank, then there are infinitely
many values of fourth power-free integers Q which are not squares and such that
G(a, Q) ⊂ Vf .
Proof. It is clear that we are interested in finding solutions of the following system
of Diophantine equations
y12 − x31 y22 − x32 y32 − x33
(2) = Q, = Q2 , = Q3
ax1 ax2 ax3
in variables xi , yi , i = 1, 2, 3 and Q. To find a solution we make the following
substitution
(3) x1 = v 2 , y1 = y, x2 = pQ, y2 = qQ2 , x3 = rQ2 , y3 = sQ3
which reduces the system (2) to the system
(4) Q = g1 (v, y) = g2 (p, q) = g3 (r, s),
where
y2 − v6 p3 + ap ar
g1 (v, y) = , g2 (p, q) = , g3 (r, s) = .
av 2 q2 s2 − r3
We take r = ap(p2 + a) and observe that the equation g2 (p, q) = g3 (r, s) defines
a quadric in variables (q, s) which contains rational points at infinity and can be
parameterized in the following way
a3 p3 u2 − a3 − 3a2 p2 − 3ap4 − p6 a3 p3 u2 + a3 + 3a2 p2 + 3ap4 + p6
q=− , s= .
2au 2u
8 MACIEJ ULAS

With q, r, s chosen in this way we left with one equation g1 (v, y) = g2 (p, q(p, u)).
If we treat p, v as a rational independent variables, then this equation defines the
curve in the plane (u, Y ), over the field Q(p, v), of the form
Ca : (a3 p3 u2 − (p2 + a)3 )2 y 2 =
a6 p6 v 6 u4 − 2a3 p(p2 + a)(p2 (p2 + a)2 v 4 − 2)v 2 u2 + (p2 + a)6 v 6 =: Fa (p, v, u).
As a consequence of our reasoning, we see that each point (u, Y ) lying on the curve
Ca , together with corresponding values of q, r, s defined above, lead to the value of
Q in the following form
4a2 p(p2 + a)u2
Q = Qa (p, u) = .
(a3 p3 u2− (p2 + a)3 )2
The genus of the curve Ca is equal to 1. Moreover, Ca contains Q(p, v)-rational
point (0, (p2 + a)3 v 3 ). Thus, the curve Ca is birationally equivalent with a curve
given by an appropriate Weierstrass equation. In fact, Ca is birationally equivalent
with the curve Ea : Y 2 = X(X − f1 (p, v))(X − f2 (p, v)) from the statement of
our theorem, via an appropriate invertible map ϕ : Ca 3 (u, Y ) 7→ (X, Y ) ∈ Ea .
We do not present the form of the map because it is quite complicated (but can
be easily computed using standard method of finding birational model of a quartic
curve of the form y 2 = Au4 + Bu3 + Cu2 + Du + q 2 with rational point (0, q), see
[11, Theorem 2.17]).
Let us observe that the discriminant of Fa with respect to u is given by
Da (p, v) = Discu (Fa (p, v, u)) = 212 a18 p10 (a + p2 )10 v 20 (p2 (p2 + a)2 v 4 − 1)2 .
Because we are interested in values of p = p0 , v = v0 such that p0 (p20 + a) is not
a square it is clear that Da (p0 , v0 ) 6= 0 and the corresponding curve Ca (p0 , v0 ) is
non-singular. This is also true for the curve Ea (p0 , v0 ).
If now p0 , v0 ∈ Q are chosen in such a way, that f1 (p0 , v0 )f2 (p0 , v0 ) 6= 0 and
the elliptic curve Ea (p0 , v0 ) obtained from Ea after specialization p = p0 , v = v0 ,
has infinitely many rational points, then we get infinitely many rational values of
Q. However, to finish the proof we need to show that if Ea (p0 , v0 ) has infinitely
many rational points, then we get infinitely many essentially different values of
Q. More precisely, we prove that there is an infinite set U ⊂ Q such that for
each u0 , u00 ∈ U the quotient Qa (p0 , u0 )/Qa (p0 , u00 ) is not a fourth power. We will
construct the sequence of finite sets (Un )n∈N recursively in the following way. Let
us fix a point P0 = (X00 , Y00 ) of infinite order on Ea (p0 , v0 ) and write (un , Yn ) =
ϕ−1 (nP0 ) ∈ Ca (p0 , v0 )(Q). Let us note that the point (u, Y ) ∈ Ca (p0 , v0 ) such that
Qa (p0 , u)/Qa (p0 , u0 ) is a fourth power, lie on the variety
Wa (u0 ) : Qa (p0 , u) = Qa (p0 , u0 )T 4 , Y 2 = Fa (p0 , v0 , u),
which, as an intersection of two surfaces, defines a curve over Q. One can check
that the genus of Wa (u0 ) is equal to 5, and thus there are only finitely many u’s
which give a rational point. Thus, there is a positive integer N0 such that for all
N ≥ N0 the quotient Qa (p0 , un )/Qa (p0 , u0 ) is not a fourth power. We thus take
U0 = {u00 , u01 }, where u00 = u0 , u01 = uN0 . Let us suppose that we constructed
the set Un−1 = {u00 , u01 , . . . , u0n } ⊂ {ui : i ∈ N} with required properties, i.e.,
for each i, j ∈ {0, . . . , n}, i 6= j the quotient Qa (p0 , v0 , u0i )/Qa (p0 , v0 , u0j ) is not a
fourth power. To find u0n+1 we note that for each i ∈ {0, . . . , n} the curve Wa (ui )
has only finitely many rational solutions (u, T, Y ). Thus, there is an integer Nn+1
 GEOMETRIC PROGRESSIONS IN THE SETS OF VALUES OF RATIONAL FUNCTIONS 9

such that for all n ≥ Nn+1 there is no rational numbers T, Y such that (un , T, Y )
lies on Wa (ui ) for i = 0, . . . , n. As a consequence of our reasoning we can take
u0n+1 = uNn+1 and put Un+1 = Un ∪ {u0n+1 }. Our reasoning shows the existence of
an increasing sequence (Un )n∈N of sets with cardinality #Un = n + 1, each having
a required property. To finish the proof, it is enough to take U = limn→+∞ Un . 

It should be noted that if p0 , v0 ∈ Q are such that Ea (p0 , v0 ) has positive rank,
then the corresponding value of Q tends to be pretty large. To see this, let us
consider the following example.
Example 3.2. Let us apply Theorem 3.1 for a specific value of a. We take a = 3,
which is the smallest positive integer such that the curve E : y 2 = x3 + 3x has a
positive rank. Next, we chose p = 4, v = 2 and the resulting elliptic curve is
E3 (4, 2) : Y 2 = X(X − 21070848)(X − 21070620).
One can check that E3 (4, 2) has rank 1 and the generator of infinite part is generated
by the point P = (X, Y ) = (732246016/9, 14683034857472/27). We have that
 
4204567 1592941018808
ϕ−1 (P ) = (u, y) = ,− .
4146944 199115595007
Performing all necessary computations we found the value of Q given by
   2
4204567 476639376
Q = Q3 4, = 19 · .
4146944 199115595007
Then we have the point Pi of infinite order on Ei (3, Q), i = 0, 1, 2, 3, where P0 =
(1, 2) and
 
1592941018808
P1 = 4, ,
199115595007
 
17266067201278872576 78182031219520468152777449472
P2 = , ,
1991155950072 1991155950073
57 · 86330336006394362882 127652133024668050546 · 517982016038366177282
 
P3 = , .
396470201749916433300492 396470201749916433300493
Remark 3.3. Let us note that the smallest positive integer Q > 1 such that (3, Q)
is a proper solution of Problem 1.1 is Q = 17. The point Pi is of infinite order on
Ei (3, 17), i = 0, 1, 2, 3, where
   
121 8206 49 102830
P0 = (1, 1), P1 = (25, 130), P2 = , , P3 = , .
25 125 121 1331
Let
R = {a ∈ N+ : rank of the curve y 2 = x3 + ax is positive}.
We performed small computer search and found that for each a ∈ R ∩ [1, 103 ],
there are p0 , v0 ∈ N such that the curve Ea (p0 , v0 ) has positive rank. In fact, in the
considered range of a, we can always take p0 from the set {2, 3, 4} and v0 form the
set {1, . . . , 10}. Although small, our computations suggest the following:
Conjecture 3.4. For each a ∈ R there is a specialization p = p0 , v = v0 such that
p0 (p20 + a) is not a square and the rank of the elliptic curve Ea (p0 , v0 ) is positive.
10 MACIEJ ULAS

In light of Theorem 3.1 one can investigate the following related question. Let
Q ∈ Z \ {−1, 0, 1} be given. Is it possible to find an a ∈ Z which is a fourth power-
free and the elliptic curve Ei (a, Q) has positive rank for i = 0, 1, 2, 3? Equivalently,
we are interested in the solutions of the system od Diophantine equations
f (x1 , y1 ) f (x2 , y2 ) f (x3 , y3 )
(5) a = f (x0 , y0 ) = = = .
Q Q2 Q3
The common value of the fractions f (xi , yi )/Qi , i = 0, 1, 2, 3 is the value of a we are
looking for. Let us put xi = pi T, yi = qi T and solve the equation f (xi , yi )/Qi =
f (xi+1 , yi+1 )/Qi+1 with respect to the variable T , for i = 0, 1, 2. As a result, we
see that to solve (5), the value of T need to be of the form
p0 q12 − Qp1 q02 p1 q22 − Qp2 q12 p2 q32 − Qp3 q22
T = = = .
p0 p1 (p21 − Qp20 ) p1 p2 (p22 − Qp21 ) p2 p3 (p23 − Qp22 )
We arrived to the following conclusion: if (xi , yi ), i = 0, 1, 2, 3 solve the system (5),
then the corresponding values of pi , qi , i = 0, 1, 2, 3, satisfy the system
(6) (
0 p1 p2 Q(p21 Q − p22 )q02 + p0 p2 (p22 − p20 Q2 )q12 + p0 p1 (p20 Q − p21 )q22 = 0,
CQ :
p2 p3 Q(p22 Q − p23 )q12 + p1 p3 (p23 − p21 Q2 )q22 + p1 p2 (p21 Q − p22 )q32 = 0,
and the corresponding value of a has the form
(p31 q02 − p30 q12 )(p1 q02 Q − p0 q12 )
a=− .
p20 p21 (p21 − p20 Q)2
The variety C 0 can be seen as a projective variety defined over the function field
K = Q(Q, p0 , p1 , p2 , p3 ). This is just an intersection of two quadratic surfaces, and
thus, from geometric point of view, we get a curve of genus 1. Unfortunately, it
seems that there is no general point on C 0 and it is non-trivial problem to find
a specialization p = (p0 , p1 , p2 , p3 ) such that the curve C 0 (p) has infinitely many
rational points and (a, Q) is a proper solution of Problem 1.1.
Example 3.5. Let Q = 2. We know that (47, 2) is a proper solution of Problem
1.1 for f (x, y) = (y 2 − x3 )/x. Using points from Table 1. one can take p =
(289/25, 14, 96, 4716544/18225)) and note that the curve C20 (p) contains the point
S = (−5712/125, 14, 96, 10271916928/2460375). Using this point as a point at
infinity, we get that C20 (p) is birational, via an appropriate map χ, with the elliptic
curve with minimal Weierstrass equation E20 (p) : y 2 = x3 + ax + b, where
a = −114789376213793508149518054796962396588,
b = −154452043874358875364810179220092982963254111141598049488.
The curve E20 (p) contains the point of infinite order
 
32968008403251534251 397625982502899147956961689375
χ(S 0 ) = − , ,
9 27
where S 0 = (5712/125, 14, 96, 10271916928/2460375). Thus, one can compute the
points χ−1 (mχ(S 0 )) for m ∈ N+ , m 6= 2, and get corresponding points on C20 (p), and
an appropriate value of am (p, q). For m = 1 we get nothing new, i.e., a1 (p, q) = 47u4
for some u ∈ Q. For m = 3, the value am (p, q) is, up to fourth power, an integer
with 1564 digits.
 GEOMETRIC PROGRESSIONS IN THE SETS OF VALUES OF RATIONAL FUNCTIONS 11

As we will see in the next section, there are many a’s such that for given Q ∈
{2, . . . , 12} the pair is a proper solution of Problem 1.1.

4. Computational observations, questions and conjectures

In this section we collect results of our computations and formulate some ques-
tions and conjectures which may stimulate further research. To compute ranks of
several thousands of elliptic curves we used very fast procedure ellrank imple-
mented in PARI/GP [7]. The function ellrank attempts to compute the rank of
a given elliptic curve E. The function returns the vector [r, R, s, L]. The rank of
E(Q) is between r and R (both included), s is related to the Tate-Shafarevitch
group, and L is a list of independent, non-torsion rational points on the curve. To
init the curve we need to use procedure ellinit. Here is a sample session:
gp> E=ellinit([47,0]);
gp> ellrank(E)
[1, 1, 0, [[289/25, 5712/125]]]
So the rank of an elliptic curve E : y 2 = x3 + 47x is one and an infinite part
of E(Q) is generated by the point (289/25, 5712/125). However, in some cases
PARI/GP is not able to compute the rank exactly, like in case of the elliptic curve
H : y 2 = x3 + 257x:
gp> H=ellinit([257,0]);
gp> ellrank(H)
[0, 2, 0, []]
We get that the rank of H(Q) is between 0 and 2. However, in such cases we
invoke the procedure ellanalyticrank which computes the pair [r, d], where r is
an analytic rank of an elliptic curve and d the value of d-th derivative at 1 of a
L-function associated to H. We have
gp> analyticrank(H)
gp> [0, 7.4090738601929029389865884843731744616]
Thus, invoking Kolyvagin’s [4] together with Wiles’ [12] result on modularity of
elliptic curves over Q, we know that the rank of H is equal to 0.
After this digression, let us back to Problem 1.1 with f (x, y) = (y 2 − x3 )/x. We
performed additional computations and looked for positive integers a, Q such that
G(a, Q) ⊂ Vf with Q ∈ {2, . . . , 12} and a ≤ 105 . For a given fourth power-free
Q ∈ N one can define the set

CQ := {a ∈ N : G(a, Q) ⊂ Vf }

and its counting function

CQ (x) = #(CQ ∩ [1, x]).

Because (a, Q) is a solution of Problem 1.1 if and only if (a, Q3 ) is a solution, we

get that for any square-free Q we have equality of sets CQ = CQ3 . Moreover, from
the first part of Theorem 2.9 applied with d = 3, we know that CQ2 is infinite for
each Q.
12 MACIEJ ULAS

For Q ∈ {1, . . . , 14} we have the following initial elements of the sets CQ (we
omit C8 due to the equality C8 = C2 ):
Q CQ
2 47, 69, 77, 79, 89, 94, 127, 138, 154, 155, 158, 171, 178, 188, 205, 219, 223, . . .
3 20, 31, 35, 37, 40, 47, 55, 60, 61, 73, 79, 92, 93, 95, 105, 111, 120, 127, 136, . . .
4 5, 14, 15, 20, 21, 31, 34, 37, 39, 46, 47, 49, 53, 55, 56, 60, 65, 66, 69, 73, . . .
5 8, 18, 19, 24, 29, 33, 40, 56, 79, 88, 90, 95, 98, 99, 104, 120, 126, 128, 129, . . .
6 15, 19, 33, 39, 40, 69, 73, 83, 85, 88, 90, 93, 95, 98, 104, 111, 114, 115, 129, . . .
7 51, 55, 67, 68, 73, 85, 89, 92, 113, 115, 120, 129, 136, 149, 155, 158, 179, . . .
9 14, 19, 20, 21, 24, 31, 34, 35, 37, 39, 40, 46, 47, 55, 60, 61, 65, 66, 67, 69, . . .
10 15, 24, 31, 34, 39, 65, 66, 69, 77, 89, 104, 111, 114, 129, 141, 143, 150, 156, . . .
11 15, 18, 20, 24, 28, 31, 47, 53, 69, 79, 84, 95, 98, 104, 111, 113, 127, 133, 136, . . .
12 14, 34, 37, 46, 47, 61, 69, 85, 92, 94, 95, 111, 126, 143, 148, 154, 157, 158, . . .
Table 2. Initial elements of the sets CQ for Q = 2, . . . , 12, Q 6= 8.
In the light of our computations one can ask the following
Question 4.1. What is the order of magnitude of CQ (x)?
Our numerical computations suggest that, at least for Q ∈ {2, . . . , 12}, the func-
tion CQ (x)/x has a limit (see picture below), say L(Q). The conjectured estimations
of L(Q) can be easily read of from the values of CQ (105 ), which are as follows
Q 2 3 4 5 6
CQ (105 ) 13623 17766 37872 17130 17456
Q 7 9 10 11 12
CQ (105 ) 16120 38562 13458 17682 17292
Table 3. Values of CQ (105 ) for Q = 2, . . . , 12, Q 6= 8.
In particular, L(2) ≈ 0.13623, L(3) ≈ 0.17766, L(4) ≈ 0.37872 and so on.
Results of our computations strongly suggest the following:
Conjecture 4.2. Let f (x, y) = (y 2 − x3 )/x.
(1) For each Q ∈ N≥2 the set CQ is not empty.
(2) For each Q ∈ N≥2 the set CQ is infinite.
Remark 4.3. From the first part of Theorem 2.9 applied for d = 3 we know that
CQ2 is infinite. The method of proof can be used to get a lower bound for CQ2 (x).
Indeed, to make things explicit as much as possible, we take in the proof of Theorem
2.9
3 1
p0 = , p1 = , q0 = 3(3Q2 − 1)(3Q2 + 1)u, q1 = (1 − 3Q2 )(3Q2 + 1)v.
4 4Q
Then, the corresponding value of a is given by
a = a(u, v) = 768Q3 (3Q2 − 1)2 (3Q2 + 1)2 (3Qu2 − v 2 )(3Q3 v 2 − u2 ).
Now, if we fix u ∈ N, then for each integer
 
1 p
v∈ √ u, 3Qu
Q 3Q
we have a(u, v) > 0. It is possible that for some u, v ∈ N+ the value a(u, v) is
a fourth power. However, the number of such possibilities is finite because the
 GEOMETRIC PROGRESSIONS IN THE SETS OF VALUES OF RATIONAL FUNCTIONS 13

(projective) curve z 4 = a(u, v) is of genus 3, and has only finitely many primitive
solutions. As a consequence we get that
CQ2 (x) > c(Q2 )x1/4
for some computable constant c(Q2 ). However, this bound is far weaker than our
expectations.

Figure 1. Plots of the functions CQ (x) for Q = 2, . . . 6 (left) and

Q = 7, . . . , 12, Q 6= 8, (right) and x ≤ 105 .

Figure 2. Plots of the functions CQ (x)/x for Q = 2, . . . 6 (left)

and Q = 7, . . . , 12, Q 6= 8, (right) and x ≤ 105 .

It is also quite interesting to note that

12
\
CQ ∩ [1, 105 ] = {32504, 34023, 36333, 40399, 41080, 41415, 55260,
Q=2
55965, 73242, 86415, 99342},
i.e., for each Q ∈ {2, . . . , 12}, i ∈ N and each a from the above set, the elliptic curve
Ei (a, Q) : y 2 = x3 + aQi x has positive rank. An additional check confirms that
13
\
CQ ∩ [1, 105 ] = {32504, 34023, 36333, 40399, 41080, 55260, 73242, 99342},
Q=2
14
\
CQ ∩ [1, 105 ] = {40399, 73242, 99342}.
Q=2
T15
However, the set Q=2 CQ ∩ [1, 105 ] is empty. Our computations suggest the fol-
lowing:
TN
Question 4.4. Let N ∈ N+ be given. Is the set Q=2 CQ non-empty?
14 MACIEJ ULAS

We expect that the answer on the question above is YES. On the other hand,
we have the following:
T∞
Proposition 4.5. We have Q=2 CQ = ∅.
T∞
Proof. Let us suppose that the statement is not true and take a ∈ Q=2 CQ .
Without loss of generality we can assume that a is fourth power-free. Write a =
q1 q22 q33 , where q1 , q2 , q3 are square-free pairwise co-prime integers. Let us take Q =
q13 q22 q3 and observe that a 6∈ CQ . Indeed, we note that the curve E1 (a, Q) : y 2 =
x3 + aQx is isomorphic over Q with the curve E : y 2 = x3 + x. However, the rank
of E(Q) is zero and the same is true for E1 (a, Q)(Q) - a contradiction. 

Next, we note that for given Q1 , Q2 ∈ {2, . . . 11}, Q1 6= Q2 , our computations

suggest that for sufficiently large x, the difference CQ1 (x) − CQ2 (x) has a constant
sign, and in fact there is CQ1 (x) − CQ2 (x) → ±∞ with x going to infinity.
In the light of our computations one can formulate the following general problem.

Problem 4.6. Let Q = {Q ∈ N+ : Q is fourth power free}.

(1) Characterize pairs Q1 , Q2 ∈ Q, Q1 > Q2 of positive integers such that the
sign of the function CQ1 (x) − CQ2 (x) is eventually constant.
(2) Does there exist Q1 , Q2 ∈ Q such that CQ1 (x) − CQ2 (x) changes the sign
infinitely often?

In the light of Theorem 3.1, for given a ∈ R one can define the quantity
m(a) = min{Q ∈ N≥2 : (a, Q) is a proper solution of Problem 1.1}.
Although it is not clear that the function is well defined, results of our computations
suggest that this is the case. However, as we observed, the corresponding value of Q
computed from rational point on the elliptic curve Ea (p, v) constructed in Theorem
3.1 is much bigger than it is expected. More precisely, from our computations it
follows that for each positive a ∈ R ∩ [1, 104 ], the value of m(a) exists. The values
of m(a) for a ∈ R ∩ [3, 60] are given in Table 4 below.

a 3 5 8 9 13 14 15 18 19 20 21 24 28 29 31
m(a) 17 17 5 17 15 12 6 5 5 3 14 5 11 5 3
a 33 34 35 37 39 40 46 47 48 49 51 53 55 56 60
m(a) 5 10 3 3 6 3 12 2 17 17 7 11 3 5 3

Table 4. The values of m(a) for a ∈ R ∩ [3, 60].

In fact, we have
{m(a) : a ∈ R ∩ [1, 104 ]} = {2, 3, 5, 6, 7, 10, 11, 12, 13, 14, 15, 17, 21, 22, 23, 26}.
Let us write R = {a1 , a2 , . . .}, where ai < ai+1 for i ∈ N+ . On Figure 4 below, we
present the plot of the function m(an ) for n ∈ {1, . . . , 6005}, i.e., an ∈ R ∩ [1, 104 ].

In the table below, for each element Q of the above set we present the smallest
value of a ∈ R such that m(a) = Q.
 GEOMETRIC PROGRESSIONS IN THE SETS OF VALUES OF RATIONAL FUNCTIONS 15

Figure 3. Plot of the function f (n) = m(rn ), where n ≤ 6005.

m(a) 2 3 5 6 7 10 11 12
a 47 20 8 15 51 34 28 14
m(a) 13 14 15 17 21 22 23 26
a 63 21 13 3 594 6295 2028 7255
Table 4. The smallest value of a such that m(a) = Q, where
Q ∈ {m(a) : a ∈ R ∩ [1, 104 ]}.
We formulate the following:
Conjecture 4.7. The function m : R → {Q ∈ N+ : Q is fourth power-free} is
well defined.
It is clear that Conjecture 3.4 implies Conjecture 4.7. However, it is not clear
whether there is an implication in the other side.
Conjecture 4.8. The function m : R → {Q ∈ N+ : Q is fourth power-free} is
onto and each element in its image is a limit point. In particular
lim inf m(a) = 2, lim sup m(a) = +∞.
a∈R a∈R

One can also ask about average behaviour of the function m(an ). On Figure 5
we plot the function P
i≤x m(ai )
A(x) =
x
for x ≤ 6005. Based on our computations we dare to formulate the following:
Conjecture 4.9. If Conjecture 4.7 is true, then the function A is bounded.
16 MACIEJ ULAS

Figure 4. Plot of the function A(x), where x ≤ 6005.

In the light of our findings concerning the questions around Problem 1.1 for
f (x, y) = (y 2 − x3 )/x, one can ask and investigated similar questions in the case of
the polynomial f (x, y) = y 2 − x3 , both form computational and theoretical point
of view. To be more precise, we are interested in the existence of proper solutions
of Problem 1.1, i.e., Q is neither as quare nor a cube and for each i ∈ N the rank
of elliptic curve
Ei0 (a, Q) : y 2 = x3 + aQi
is positive. We show that (3, 2) is a proper solution of Problem 1.1. To see this,
one can compute the following data:
i r(Ei0 (3, 2)) generators of Ei0 (3, 2)(Q)
0 1 (−2, 5)
1 1 (1/4, 65/8)
2 1 (4, 14)
3 1 (−2, −16)
4 2 (16, 68), (−8, 4)
5 1 (48217/5041, −15728083/357911)
Table 5. Ranks and the generators for the elliptic curves
Ei0 (3, 2) : y 2 = x3 + 3 · 2i , i = 0, . . . , 5.
Consequently, the pair (32, 2) is a proper solution of Problem 1.1 for f (x, y) =
y 2 − x3 .
We formulate the following:

Problem 4.10. Investigate properties of the set of proper solutions of Problem 1.1
for f (x, y) = y 2 − x3 . In particular prove the following:
 GEOMETRIC PROGRESSIONS IN THE SETS OF VALUES OF RATIONAL FUNCTIONS 17

(1) For each a ∈ Z such that the elliptic curve y 2 = x3 + a has positive rank,
there are infinitely many values of Q ∈ Z such that (a, Q) is a proper
solution of Problem 1.1.
(2) For each Q ∈ Z \ {−1, 0, 1}, there are infinitely many values of sixth power-
free integers a such that (a, Q) is a proper solution of Problem 1.1.

References
[1] W. Bosma, J. Cannon, C. Playoust, The Magma algebra system. I. The user language. J.
Symbolic Comput., 24(3-4) (1997), 235–265.
[2] F. Beukers, On the generalized Ramanujan-Nagell equation I, Acta Arith. 38(4) (1981), 389–
410.
[3] Y. Bugeaud, T. N. Shorey,On the number of solutions of the generalized Ramanujan-Nagell
equation, J. reine angew. Math. 539 (2001), 55–74.
[4] V. Kolyvagin, Finiteness of E(Q) and X(E, Q) for a class of Weil curves, Math. USSR Izv.
32 (3) (1989), 523–541.
[5] K. Mahler, Zur Approximation algebraischer Zahlen. I, Math. Ann. 107(1) (1933), 691–730.
[6] T. Nagell, The Diophantine equation x2 + 7 = 2n , Ark. Mat. 30 (2-3) (1961), 185–187.
[7] The PARI Group, PARI/GP version 2.15.2, Univ. Bordeaux, 2022, http://pari.math.
u-bordeaux.fr/.
[8] S. Ramanujan, Question 464, J. Indian Math. Soc. 5 (1913), 130.
[9] T. N. Shorey, R. Tijdeman, Exponential Diophantine equations, Cambridge Tracts in Math-
ematics. Vol. 87. Cambridge University Press 1986.
[10] A. Thue, Über Annäherungswerte algebraischer Zahlen, J. Reine Angew. Math., 135 (1909),
284–305.
[11] L. C. Washington, Number Theory and Cryptography, Discrete Mathematics and its Appli-
cations. Boca Raton, FL: Chapman and Hall/CRC 2003.
[12] A. Wiles, Modular elliptic curves and Fermat’s last theorem, Ann. Math. 141 (3) (1995),
443–551.
[13] Wolfram Research, Inc., Mathematica, Version 13.2. (Champaign, IL, 2022).