Elvira LupoianandJames Rawson
Abstract.
In this paper we give an algorithm to find the 3-torsion subgroup of the Jacobian of a smooth plane quartic curve. We describe torsion points in terms of cubics which triply intersect the curve, and use this to define a system of equations whose solution set corresponds to the coefficients of these cubics.We compute the points of this zero-dimensional, degree scheme first by approximation, using hom*otopy continuation and Newton-Raphson, and then using lattice reduction or continued fractions to obtain accurate expressions for these points.We describe how the Galois structure of the field of definition of the -torsion subgroup can be used to compute local conductor exponents, including at .
Key words and phrases:
Torsion, Jacobians, Plane Quartics, Conductors, LLL, hom*otopy Continuation
2000 Mathematics Subject Classification:
11Y60 (primary), 11G30, 11Y50
Both authors are supported by the Warwick Mathematics Institute Center for Doctoral Training, and gratefully acknowledge the funding from the UK Engineering and Physical Sciences Research Council. The first named author is supported by the grant EP/V520226/1. The second named author is supported by the grant EP/W523793/1.
1. Introduction
The celebrated result of Mazur [17] gives a complete classification of rational torsion subgroups of elliptic curves. The general study of torsion of abelian varieties is considered out of reach. An interesting variant of this question could be the study of torsion points, for a fixed natural number .For Jacobians of genus 1 curves, this is a largely elementary problem, since explicit equations are known for the Jacobian (as it is an elliptic curve) and the group law is accessible. For Jacobians of higher genus curves, the problem becomes much more difficult. Perhaps the clearest description is that of torsion, since for the hyperelliptic curves, it can be easily deduced from the Weierstrass model, and for non-hyperelliptic curves -torsion points correspond to multi-tangent hyperplane to the curve, as described in [11]. In [5], an explicit description of torsion is derived for Jacobians of genus curves. This is generalised in [15], for hyperelliptic curves of genus . For hyperelliptic curves, such description of torsion can be obtained using a Weierstrass model for the curve and careful analysis of Riemann-Roch spaces. However, when the curve is not hyperelliptic, the description of torsion points tends to become more geometric.
Explicitly computing such torsion subgroups has important applications. For instance, it’s used for descent purposes in [5] and [6]; and also for conductor computations in [10].
In this note, we give a description of the 3-torsion of genus 3, non-hyperelliptic curves in terms of degree 3 curves meeting the original quartic in 4 places, each with multiplicity 3. As a consequence of this description, we are able to give equations for all of the unramified covers of the curve with Galois group .
After rigidifying the configuration of a cubic and the genus 3 curve, we give a schema for calculating all such curves. Using numerical methods (hom*otopy continuation and Newton-Raphson), we find high-precision complex approximations to the cubics. The coefficients can be recognised as algebraic integers by algorithms like Lenstra-Lenstra-Lovász (LLL) or continued fraction methods on the minimal polynomial. The 3-torsion subgroup computed is then verified by reduction modulo primes. A similar method of computing torsion is described in [16] and [15].
We conclude by providing detailed examples in the cases of the Fermat quartic, , and the Klein quartic, . For these curves we compute the conductor exponents by torsion methods, and compare against the value computed from their structure as modular curves. Symbolic calculations were carried out using the Magma software package [3] and numerics with Julia [1].
2. Derivation
From now on, will be a genus , non-hyperelliptic, smooth, projective, geometrically integral curve, embedded into via the canonical embedding. Denote by the Jacobian variety attached to . We note that is isomorphic to the Picard group, and we regard its points as equivalence classes of degree zero divisor on the curve. Let be a 3-torsion point of the Jacobian. As is degree 0, , so after fixing a choice of canonical divisor, , can be written as , where both and are effective. By assumption, , so . Sections of are cubics, and so is the intersection of with a plane cubic. As is the triple of an effective divisor, the cubic must meet with multiplicity (a multiple of) 3 at every intersection point.
The divisor has degree 4, and so moves (). We may, therefore, assume contains a flex. By change of coordinates, it can further be assumed this flex is at with tangent line . Functions vanishing to order 3 at this point lie in the ideal , and so the class of cubics to be considered is those in long Weierstrass form (in the affine chart ).
Not all cubics of this form give rise to non-trivial torsion elements, however. If , then is a canonical divisor and so the intersection of with a line, and it contains . The only lines through this point are the lines of constant , hence cubics that are must be ruled out. One way to do this is to assume that at least one coefficient of is non-zero, although this may exclude some desired cubics.
If is not a special divisor, then this representation is unique; if , then , and so there is equality of divisors as neither moves. As the divisors are the same, the cubics cutting them out must be the same up to scaling.
If is a special divisor, then it can be written as , for some canonical divisor and point . It follows that, up to linear equivalence, , and so . This shows is a flex. The tangent to meets at another point, , and the divisors equivalent to are the intersection of with lines through . As , the tangent to must meet again at . So as long as has a flex whose corresponding flex line does not meet another in this way, this method will compute all non-trivial three torsion, and find unique expressions for each element.
This method can be adapted for not a flex, for instance if such a flex does not exist or to make use of a rational point. In this case, the cubics to be considered are those of the form , where is a constant depending on the curvature of at .Thus we have proved the following.
Theorem 1.
Let be a smooth, projective, geometrically integral, non-hyperelliptic curve of genus , defined over a number field and canonically embedded in . Assume that has a flex at , with tangent line , then the torsion points of it’s Jacobian correspond to cubics of the form
which intersect the curve with multiplicities divisible by . Moreover, this correspondence preserves the Galois action.
This description of -torsion points allows us to derive equations, whose solution set represents coefficients of cubics of the above form.
2.1. Scheme of -torsion points
Let be a curve as above, with an affine chart given by . Suppose that cubics corresponding to -torsion points on the Jacobian are of the form
and we assume that at least one of is non-zero. We describe the case when and derive a system of equations whose solutions parameterise . Other cases are very similar.
We want to find such that the plane cubic defined by
intersects the curve at points, each with multiplicity . Observe that the above gives an expression , and the monomials of are not divisible by . The equation defining our affine quartic can be viewed as an equation in , and it’s of the from
with . The intersection of the two is be described by
| | | |
| | | |
for some . We repeat this to eliminate the monomial in the above expression to obtain
and hence the above gives us an expression for . Substituting for in the cubic gives us a degree polynomial , and to ensure that the points of the intersection of our cubic and quartic occur with multiplicity , we require to be the cube of a degree polynomial in , that is,
and equating the coefficients of the above expression gives equations in . Moreover, to avoid singular solutions, we require that and have no common factors, which can be imposed by adding the equation
to our system, where denotes the resultant of the two polynomials.
Theorem 2.
Let be a curve as in theorem 1. The subgroup of torsion points of its Jacobian is explicitly computable by a finite algorithm.
Theoretically, Gröbner basis techiques can be used to solve the system of equations outlined above. However, due to the degree of the scheme, this methodology is impractical. Instead, we employ a two step process. Firstly, we compute high precision complex approximations, using hom*otopy continuation and Newton-Raphson. We then obtain algebraic expressions for the points via lattice-reduction or continued fraction methods. This is outlined in the following sections.
3. Approximations and hom*otopy Continuation
Let be equations defining a zero-dimensional scheme, the points of which give the coefficients of our cubics. In this section we give an overview of how to obtain high precision approximations using the classical Newton-Raphson method and hom*otopy continuation. We begin with a brief summary of the Newton-Raphson method, a comprehensive overview can be found in [21, Page 298].
Consider the valued function defined by ,
.
Let be the Jacobian matrix of and suppose is an approximate solution of with invertible.
For each integer , define
.
If the initial approximation is good enough, the resulting sequence converges to a root of , with each iterate having increased precision. For our computations, we efficiently implemented Newton-Raphson in Julia.
We compute the required initial approximations using hom*otopy continuation and its implementation in Julia (see [4]).
3.1. hom*otopy Continuation
hom*otopy continuation is a method for numerically approximating the solutions of a system of polynomial equations by deforming the known solutions of a similar system. In this subsection we give a sketch of the idea, a detailed explanation of this theory can be found in [23] .
We denote by , where is the maximum degree of the monomials of . Let be a system of polynomials in , with exactly solutions, all of which are known or easily computable. This will be known as a start system. The standard hom*otopy of and is a function
| | |
| | |
Fix , and define for . For large enough, the solutions of are good approximations of the solutions of , and their precision can be increased by using Newton-Raphson. The solutions of are known, and they define paths to approximate solutions of .
Notably, given any system , a start system (and its solutions) can always be computed. hom*otopy continuation is efficiently implemented in the Julia package hom*otopyContinuation.jl (see [4]). This implementation gives approximates solutions which are accurate to 16 decimal places. For our computations we used the approximate solutions and 200 iterations of Newton-Raphson to obtain an accuracy of 600 decimal places.
4. Lattice-Reduction and Continued Fractions Techniques
Let be a point on a scheme for which we have a complex approximation of arbitrarily large precision .
One way of computing minimal polynomials is by searching for short vectors in an appropriately defined lattice. Short vectors can be computed using the Lenstra-Lenstra-Lovász lattice reduction algorithm (LLL) which, given a lattice, returns a reduced basis whose vectors have small norms, see [7, Section 2.6]. This is a standard technique for computing short vectors, for instance see [7, Section 2.7.2] or [19, Chapter 6.1]. To compute an LLL- basis we use the Julia implementation provided by LLLplus.jl.
When computing the shortest vector in an appropriately defined lattice is impractical, we can also attempt to find the minimal polynomial by using our approximations and continued fractions to compute the -rational polynomial whose roots are all of the th coordinates of points on the scheme.
4.1. Lattice Reduction and Minimal Polynomials
We give an overview of the theory required to compute our minimal polynomials. This was necessary in our computations as we made use of the Julia’s LLL implementation, which was significantly faster then using Magma’s built in commands, such as MinimalPolynomial().
Fix and let and . We want to find and such that
.
We assume that the imaginary part of is very small, and hence we are probably approximating a real number . At the end of this section, we explain how the method described below can be adapted when is not real.
Let and fix a constant , for some large natural number , such that
for all
where denotes the integer part of a real number . Let be the lattice generated by the columns of the matrix
Define and observe that
The vector of coefficients of the required polynomial can be recovered from , by setting
.
The norm of is bounded by the norm of as follows,
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
and hence although depends on the precision of the approximation, it is bounded by the fixed constant .
As explained in [19, Page 68], heuristically the length of the shortest vector in is approximately , where is the determinant of the lattice . Thus, for a sufficiently large precision, the vector is expected to be the shortest vector in the lattice, and of norm significantly smaller than the stated bound. In our case, ; and so if the minimal polynomial has coefficients of order , , are such that:
then short vectors in are a suitable candidates for the vector of coefficients.
When the imaginary part of is not small, our strategy is to simultaneously minimise both real and imaginary parts of the polynomial, and hence the same method can be applied to the lattice generated by the columns of
where and denote the real and imaginary parts of ; and where , is such that
and
for all .
Note that the above construction depends on , which we don’t a priori know. However, we can replace in the construction by any positive integer and search for short vectors in the corresponding lattice. The strategy to find is as follows; we begin with a candidate for , starting with , and running through the natural numbers. If our candidate is equal to , then the norm of the shortest vector is typically much smaller than the expected bound, and we identify it as a possible vector of coefficients.
Coefficient Relations
Given minimal polynomials of , we want to find which combinations of roots correspond to points on the scheme.
We do this by searching for relations amongst our coefficients using lattice reduction. We show how a possible relation between and is computed. If , we search for such that
where is the degree of the minimal polynomial of . As in our search for minimal polynomials, such integers can be found by computing short vectors in the lattice generate by the columns of
where and is chosen similarly as before. If the imaginary parts of are not both small, we instead search for short vectors in the lattice generated by the columns of
In all of our examples, for a given orbit of cubics, we were able to express all coefficients of the defining equations in terms of one fixed coefficient. Using these relations and the minimal polynomials, we verify their correctness by computing symbolically the roots and the corresponding points on the scheme, and verifying that these are solutions of the system of equations.
4.2. Continued Fractions and Minimal Polynomials
An alternative method for determining minimal polynomials is to use the theory of continued fractions. Let be complex approximations to the roots of a rational monic polynomial of degree . Using Vieta’s formulae [13], we can construct a monic polynomial of degree whose roots are the . This polynomial will be an approximation to the original polynomial, and so it suffices to recognise the coefficients as rational numbers. The error propagation in this process is fairly tame, to first order, the relative error in the estimates is magnified by a factor of in computing the coefficients. To see this, let the relative errors in each be , then the absolute error in the coefficient of is the following.
| | |
| | |
The double sum is then the coefficient counted times, once for each choice of .Unfortunately, truncation errors become an issue if there are roots of extreme sizes, so this method tends to work best when roots are all of a similar order of magnitude.
We focus now on recognising the coefficients as rationals. This method is standard in many undergraduate lecture notes, including [20]. Given a real number, , the “best” rational approximation comes from the continued fraction expansion. In this case, is a close approximation to a rational number, . The continued fraction for terminates, and so the continued fraction expansion for will, at some point, have a very small error, corresponding to the end point of the continued fraction. If and , the error at this step is approximately , so for any rational, if is known with enough precision, the error will be below any given bound.
It is possible to factorise the resulting polynomial, and so compute Galois orbits. Our main use of this is to identify small orbits where lattice reduction techniques can be applied. As before, the correctness can be verified by computing symbolically the roots.
5. Example 1: Fermat Quartic
We work with the following affine model of the Fermat quartic
to ensure that a flex is in the necessary configuration.The -torsion points on the Jacobian of are effective divisor , where is the zero divisor of ,
,
for some algebraic numbers .
Following the method described in Section 2, we arrive at the following ten equations:
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
We also require an additional eleventh equation:
where is the resultant of two polynomials, as previously described.
Using the method described in the previous sections, we compute two orbits of solutions. We state our orbits as follows: we give the minimal polynomial of , and expressions for in terms of .
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
Another orbit of -torsion points can be obtained by working on the simpler scheme, where we assume that our cubics are of the form
,
and derive equations for this scheme by eliminating variables as in the previous case. This scheme has degree and we find that its points form a single orbit. To describe our orbit, we give a minimal polynomial for and expression for the other coefficients in terms of .
| | |
| | |
| | |
| | |
| | |
| | |
We verify that the three orbits above generate the entire subgroup as follows. The cubics above are all defined over the degree number field given by
.
Let be the finite subgroup of corresponding to the points in above orbits. In the ring of integers of , , there is a prime ideal of norm , and since is a prime of good reduction for the curve, reduction modulo induces an injection on torsion [14]. The image of under the reduction map
is , and hence is necessarily the entire -torsion subgroup .
6. Example 2: Klein Quartic
We work with the classical affine model of the Klein quartic
.
and compute such that the cubic defined by
gives a -torsion element. Then proceeding as before, we compute a system of equations whose solutions correspond to our coefficients:
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
and is of the same form as before.
We computed two orbits of torsion points, which were sufficient to generate the entire group. Below, we give the minimal polynomial of and expressions for in terms of .
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
Both of our orbits are defined over the degree number field given by
and as in the previous example, we reduce modulo a prime of good reduction to verify that the two orbits listed above generate the entire torsion subgroup .
7. Local Conductor Exponents
A interesting application of these calculations is to compute local conductor exponents.
Let be any smooth, projective curve of genus defined over . The conductor of is an arithmetic invariant of the curve, defined as a product over the primes of bad reduction for the curve. The problem of computing conductors is essentially the problem of computing the local conductor exponent at each primes of bad reduction for the curve.
For elliptic curves, Tate’s algorithm [18] gives a complete resolution of the problem. This is also completely solved in the genus case by combining the cluster picture method described in [9] for with the algorithm of Dokchitser-Doris in [10] for . In fact the results of [9] hold more generally for hyperelliptic curves and . In [15], the method of [10] was generalised to compute part of for hyperelliptic curves of genus . Following [10], we explain how the torsion subgroup of the Jacobian can be used to compute the wild part of .
Below, we give an overview of relevant definitions and results, see [22] for details.
To compute , we view as a smooth projective curve over , let be the Jacobian variety associated to and for a prime let and be the associated adic Tate module and adic representation respectively. Then is the Artin conductor of defined as follows
where are the ramification subgroups in lower numbering, defined as
The definition is independent of the choice of prime .
For computational purposes, it’s convenient to break up the above quantity into two parts. First, we observe that is the inertia subgroup, and define the ‘tame part’ of as
.
This is computable directly from the regular model of the curve over , as we can deduce the following invariants:
- •
the abelian part , equal to the sum of the genera of all components of the model,
- •
the toric part , equal to the number of loops in the dual graph of .
The tame part of the exponent is equal to , see [2, Chapter 9] for details. Regular models can be computed in principle by taking any model of the curve and performing repeated blowups until it becomes regular or by computing a -regular model as in [8].
The second part of the conductor exponent is the ‘wild part’, defined as
.
The above sum is, in fact, finite since for sufficiently large . Also, for , is pro-p and , hence the wild conductor exponent is:
.
Thus for a give curve explicitly knowing and is sufficient to determine all .
Finding for is computationally a much simpler problem; one could apply the method described above to the -torsion subgroup, for which there is an explicit description.For a plane quartic , one could explicitly compute and hence for using the bitangent lines to the curve, and for the wild conductor exponent can be determined using our explicit description of the torsion subgroup.
8. Local Conductor Exponents at of the Fermat and Klein Quartics
We set in the previous section and compute in our examples. All of our computation are done in Magma, using the ‘Local Arithmetic Fields’ implementation [3].
8.1. The Fermat Quartic
Let , where is the defining polynomial of from Section 5, and the ramification groups in lower numbering. Using Magma, we find that is generated by five elements, , where have order 3, and have order 2. The ramification groups are as follows
| | |
| | |
| | |
| | |
and . We compute the Galois action on the -torsion subgroup by computing the action of the generators on the roots of the minimal polynomials defining our -torsion orbits. We find
| | |
| | |
From the formula, we compute the wild conductor exponent at to be
.
Observe that as , since the 3-adic Tate module maps onto the 3 torsion, and so the tame part of the conductor is 6. Combining this gives the conductor exponent at 2 as .In this example, we can verify our computation using the the modular structure of the Fermat quartic. The curve is isomorphic to , and there are 3 cusp forms of this level and weight 2, two copies of the cusp form of level 32 and one of level 64. The conductors of these forms are the conductors of the corresponding elliptic curves arising in the Jacobian. Moreover, since the tame and wild parts of the conductor are preserved by the rational isogeny, the conductor exponent of the Fermat quartic is the sum of the 3 conductor exponents of the elliptic curves, which is in this case.
8.2. The Klein Quartic
We start by computing the conductor at 2. Since the number field defined by , as in Section 6, is unramified above 2, the ramification groups beyond are trivial. In particular, the wild conductor at 2 is trivial. This is to be expected, since the curve is known to have good reduction at 2, and so has trivial conductor at 2. Instead, we demonstrate the method at 7, the only bad prime for this curve. Let , where is the defining polynomial of from Section 6. We denote the ramification groups in the lower numbering by . As has order 24, which is coprime to 7, is trivial since it is a -group. This shows the wild conductor is again, trivial.
As in the previous example, we compute that , and so there are no non-trivial fixed points of the Tate module. This shows .
As in the case of the Fermat quartic, this can be deduced from the well understood theory of this curve. The Jacobian of the Klein quartic is isogenous to the cube of an elliptic curve with CM by [12]. This elliptic curve has conductor , and so . Moreover, as the conductor of the elliptic curve is all tame, the same is true of the Klein quartic, as noted above.
References
- [1]J.Bezanson, A.Edelman, S.Karpinski, and V.B. Shah.Julia: A fresh approach to numerical computing.SIAM review, 59(1):65–98, 2017.
- [2]S.Bosch, W.Lütkebohmert, and M.Raynaud.Néron models, volume21.Springer Science & Business Media, 2012.
- [3]W.Bosma, J.Cannon, and C.Playoust.The Magma algebra system. I. The user language.J. Symbolic Comput., 24(3-4):235–265, 1997.Computational algebra and number theory (London, 1993).
- [4]P.Breiding and S.Timme.hom*otopyContinuation.jl: A Package for hom*otopyContinuation in Julia.In International Congress on Mathematical Software, pages458–465. Springer, 2018.
- [5]N.Bruin, E.V. Flynn, and D.Testa.Descent via (3, 3)-isogeny on jacobians of genus 2 curves.arXiv preprint arXiv:1401.0580, 2014.
- [6]N.Bruin and M.Stoll.Two-cover descent on hyperelliptic curves.Math. Comp., 78(268):2347–2370, 2009.
- [7]H.Cohen.A course in computational algebraic number theory, volume 138of Graduate Texts in Mathematics.Springer-Verlag, Berlin, 1993.
- [8]T.Dokchitser.Models of curves over dvrs, 2021.
- [9]T.Dokchitser, V.Dokchitser, C.Maistret, and A.Morgan.Arithmetic of hyperelliptic curves over local fields, 2018.
- [10]T.Dokchitser and C.Doris.3-torsion and conductor of genus 2 curves.Mathematics of Computation, 88(318):1913–1927, 2019.
- [11]I.V. Dolgachev.Classical algebraic geometry.Cambridge University Press, Cambridge, 2012.A modern view.
- [12]N.D. Elkies.The klein quartic in number theory.In The Eightfold Way: The Beauty of Klein’s Quartic Curve.Cambridge University Press, 2001.
- [13]A.Girard.Invention nouvelle en l’algèbre.Bierens de Haan, 1884 (reprint).
- [14]N.M. Katz.Galois properties of torsion points on abelian varieties.Invent. Math., 62(3):481–502, 1981.
- [15]E.Lupoian.Three-torsion subgroups and conductors of genus 3 hyperellipticcurves, 2023.
- [16]E.Lupoian.Two-torsion subgroups of some modular jacobians, 2023.
- [17]B.Mazur.Modular curves and the eisenstein ideal.Publications Mathématiques de l’Institut des HautesÉtudes Scientifiques, 47(1):33–186, 1977.
- [18]J.H. Silverman.Advanced topics in the arithmetic of elliptic curves, volume151.Springer Science & Business Media, 1994.
- [19]N.P. Smart.The algorithmic resolution of Diophantine equations,volume41 of London Mathematical Society Student Texts.Cambridge University Press, Cambridge, 1998.
- [20]W.A. Stein.Recognizing rational numbers from their decimal expansion.https://wstein.org/edu/2007/spring/ent/ent-html/node74.html,2007.Accessed 22/01/24.
- [21]J.Stoer, R.Bulirsch, R.Bartels, W.Gautschi, and C.Witzgall.Introduction to numerical analysis, volume 1993.Springer, 1980.
- [22]D.Ulmer.Conductors of l-adic representations.Proceedings of the American Mathematical Society,144:2291–2299, 2016.
- [23]J.Verschelde.hom*otopy continuation methods for solving polynomial systems.1998.