Abstract
In a recent paper we proved some new bounded height results for equations involving varying integer exponents. Here we make a start on the problem of generalizing to rational exponents, which corresponds to the step from groups that are finitely generated to groups of finite rank. We discover two unexpected obstacles. The first is that bounded height may genuinely fail in the neighbourhood of certain exponents. The second concerns vanishing subsums, which seem to be much harder to deal with than in classical situations like S-unit equations. But for certain simple and natural equations we are able to clarify the first obstacle and eliminate the second. The proofs are partly based on our earlier work but there are also new considerations about successive minima over function fields.
Similar content being viewed by others
Data availability
The authors declare that the data supporting the findings of this study are available within the article.
Notes
That is, it has a basis of vectors in \({\overline{\mathbb {Q}}}^r\), or equivalently a set of defining linear equations over \({\overline{\mathbb {Q}}}\).
By assumption \(\mu _1\le \mu _{d_{j-1}+1}<C_3^{-1}\), which is the condition we need to apply (2).
With the convention \(\textrm{Span}_{\overline{\mathbb {Q}}}(\textbf{g}^{(1)}_P,\ldots ,\textbf{g}^{(d_0)}_P)=\{0\}\).
We denote by \(H=\exp (h)\) the affine non-logarithmic Weil’s height.
References
Amoroso, F.: On a conjecture of G. Rémond. Ann. Sc. Norm. Super. Pisa Cl. Sci. (5) 15, 599–608 (2016)
Amoroso, F., Masser, D., Zannier, U.: Bounded height problems and Silverman specialization theorem. Duke Math J. 166(13), 2599–2642 (2017)
Bérczes, A., Evertse, J.-H., Györy, K.: Effective results for linear equations in two unknowns from a multiplicative division group. Acta Arith. 136, 331–349 (2009)
Beukers, F.: On a sequence of polynomials. In Algorithms for algebra. J. Pure Appl. Algebra 117–118, 97–103 (1997)
Beukers, F., Schlickewei, F.: The equation \(x + y = 1\) in finitely generated groups. Acta Arith. 78(2), 189–199 (1996)
Bombieri, E., Masser, D., Zannier, U.: Intersecting a curve with algebraic subgroups of multiplicative groups. Int. Math. Res. Not. 20, 1119–1140 (1999)
Bombieri, E., Masser, D., Zannier, U.: Finiteness results for multiplicatively dependent points on complex curves. Mich. Math. J. 51, 451–466 (2003)
Denz, A.: Bounding the height of certain algebraic numbers, Master Thesis, University of Basel (2016)
Faltings, G.: The general case of S. Lang’s conjecture. Barsotti Symposium in Algebraic Geometry (Abano Terme, 1991), pp. 175–182, Perspect. Math., vol. 15. Academic Press, San Diego (1994)
Habegger, P.: Intersecting subvarieties of \({{\mathbb{G} }_{\rm m}}^n\) with algebraic subgroups. Math. Ann. 342(2), 449–466 (2008)
Hindry, M.: Autour d’une conjecture de S. Lang. Invent. Math. 94, 575–603 (1988)
Mignotte, M., Shorey, T.N., Tijdeman, R.: The distance between terms of an algebraic recurrence sequence. J. Reine Angew. Math. 349, 63–76 (1984)
Pottmeyer, L.: Fields Generated by Finite Rank Subgroups of Tori and Elliptic Curves. Int. J. Number Theory 17(5), 1079–1089 (2021)
Roy, D., Thunder, J.: An absolute Siegel’s lemma. J. Reine Angew. Math. 476, 1–26 (1996) [Addendum and erratum: J. Reine Angew. Math. 508, 47–51 (1999)]
Schmidt, W.M.: Diophantine approximations and Diophantine equations LNM, vol. 1467. Springer, Berlin (1991)
Silverman, J.H.: Heights and the specialization map for families of abelian varieties. J. Reine Angew. Math. 342, 197–211 (1983)
Waldschmidt, M.: Diophantine approximation on linear algebraic groups. Transcendence properties of the exponential function in several variables. Grundlehren der mathematischen Wissenschaften, vol. 326. Springer, Berlin (2000)
Warin, O.: On \(x+y+z+w=1\) and heights. Master Thesis, University of Basel (2012)
Zagier, D.: Algebraic numbers close to both 0 and 1. Math. Comput. 61, 485–491 (1993)
Zhang, S.: Positive line bundles on arithmetic surfaces. Ann. Math. 136, 569–587 (1992)
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors of the manuscript certify that they have no affiliations with or involvement in any organization or entity with any financial interest (such as honoraria; educational grants; participation in speakers’ bureaus; membership, employment, consultancies, stock ownership, or other equity interest; and expert testimony or patent-licensing arrangements), or non-financial interest (such as personal or professional relationships, affiliations, knowledge or beliefs) in the subject matter or materials discussed in this manuscript.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: Beukers’ equation with a rational exponent
In this appendix we prove the following theorem which generalizes Theorem 1.2 stated in the introduction.
Theorem A.1
Let \(\alpha \), \(\beta \) be non-zero algebraic numbers and let \(\lambda \in {\mathbb {Q}}\) be positive. Let \(t_0\in {\overline{\mathbb {Q}}}\backslash \{0,1\}\). We fix determinations of \(t_0^\lambda \) and of \((1-t_0)^\lambda \). Let us assume
and
Then
The special algebraic numbers \(t_0\) such that \(\alpha t_0^{\lambda -1}=\beta (1-t_0)^{\lambda -1}=1\) trivially satisfy the equation \(\alpha t^\lambda +\beta (1-t)^\lambda =1\). They can be directly handled as we explain after the proof of the theorem. Thus Theorem 1.2 follows from Theorem A.1; see again the discussion after the proof for details.
Our theorem extends the following result of Beukers and Schlickewei which deals with even integers \(\lambda \), and will be deduced from it.
Theorem A.2
([5, Lemma 2.3]) Let a, b, A, \(B\in {\overline{\mathbb {Q}}}^*\) such that
for some integer \(n\in {\mathbb {N}}\). ThenFootnote 4\(H(A,B)\le 6\sqrt{3}\cdot 2^{1/n}H(a,b)^{1/n}\).
Proof of Theorem A.1
Let \(\lambda >0\) be a rational parameter and let \(t_0\in {\overline{\mathbb {Q}}}\) which satisfies \(\alpha t_0^\lambda +\beta (1-t_0)^\lambda =1\) and such that
We shall prove:
Theorem A.1 follows after a simple computation.
The strategy of the proof of (A.1) is the following. We distinguish three cases: for \(\lambda \ge 6\) we apply Theorem A.2 choosing for n the integer part of \(\lambda /2\). For \(\lambda \in [1,6)\) we simply use the relations between roots and coefficients. Finally, we reduce the case \(\lambda \in (0,1]\) to the previous ones by a duality argument involving exponent \(1/\lambda \).
First case: \(\lambda \ge 6\). Let n be the integer part of \(\lambda /2\). By assumption \(n\ge 3\). Let \(a_0=t_0^{\lambda -2n}\), \(b_0=(1-t_0)^{\lambda -2n}\) and \(a=\alpha a_0\), \(b=\beta b_0\). Thus \(at_0^{2n}+b(1-t_0)^{2n}=1\). By Theorem A.2,
Since \(\lambda -2n\ge 0\),
Since \(\lambda <2(n+1)\) and \(n\ge 3\) we have
and \(2^{1/n}\le 2^{1/3}\). Thus
We have \(3\log (6\sqrt{3}\cdot 2^{1/3})\le 8\) and \(n>\lambda /2-1\ge \lambda /3\) (since \(\lambda \ge 6\)). Thus
Second case: \(1\le \lambda \le 6\). Note that we can find a rational p/q with \(\gcd (p,q)=1\) and \(q=1\) or \(q=2\) such that
(indeed, if the fractional part \(\{\lambda \}\) is \(\le 0.4\) or \(\ge 0.6\), then the inequality holds with \(q=1\); otherwise \(\vert \{\lambda \}-1/2\vert \le 0.1\) and it holds with \(q=2\)). Since \(\lambda \ge 1\) and \({\varepsilon }<1\) we have \(p\ne 0\). Let \(a_0=t_0^{\lambda -p/q}\), \(b_0=(1-t_0)^{\lambda -p/q}\) and \(a=\alpha a_0\), \(b=\beta b_0\). Thus
which means that there exists \(s_0\in {\overline{\mathbb {Q}}}\) such that \(s_0^q=t_0\) and \(1-a s_0^p=b (1-s_0^q)^{p/q}\). Taking q powers, we see that \(s_0\) is a root of the polynomial
Fact. \(f\ne 0\).
Proof
We have
Assume first \((p,q)\ne (1,1)\). Since \(\gcd (p,q)=1\) we have \(p\ne q\). We also recall that p, \(q\ge 1\). Comparing the two expansions above, we deduce that \(b^q=1\). Moreover, if \(q<p\) then \(b^q=0\), which is clearly impossible. But \(a\ne 0\) since \(\alpha \ne 0\), \(a\ne 0\). Thus \((p,q)=(1,1)\) and \(f=(1-b)+(b-a)s\) which is \(=0\) if and only if \(a=b=1\), that is
which we have excluded in our assumption. \(\square \)
We now recall a classical relation between the height of the roots of a polynomial and the height of its coefficients. Let \(K:={\mathbb {Q}}(a,b)\). Given a place v of K we denote by \(M_v(f)\) the Mahler measure of \(f^\sigma \), if v is archimedean and corresponds to the immersion \(\sigma :K\hookrightarrow {\overline{\mathbb {Q}}}\). If v is non-archimedean, we let \(M_v(f)\) be the maximum of the v-adic absolute values of the coefficients of f. We then define the normalized height of f as
where v runs over the places of K and where \(d_v\) denote the local degree. It is well known that \(\hat{h}(f)=\sum _s h(s)\) for s running over the roots of f, counted with multiplicities. Thus \({h^{\mathbb {A}}}(s_0)\le \hat{h}(f)\) and we have to estimate this last quantity.
Given a non-archimedean place of K we have
Let now v be an archimedean place, corresponding to the immersion \(\sigma :K\hookrightarrow {\overline{\mathbb {Q}}}\). Since the Mahler measure of a polynomial is bounded by the maximum of the absolute value of the polynomial on the disk of radius 1, we have:
Thus
By our choices of \(a_0\), \(b_0\) and \({\varepsilon }\),
and \(p/q\le \lambda +\epsilon \le 7\). Thus, taking into account \(q\le 2\),
and so
Third case: \(\lambda \le 1\). We reduce to the previous two cases. Note that \(t'_0=\alpha t_0^\lambda \) is a solution of
and
Since \(\lambda ^{-1}\ge 1\) the results of the previous cases apply. Suppose first \(\lambda \le 1/6\). Then \(\lambda ^{-1}\ge 6\) and
Suppose now \(1/6\le \lambda \le 1\). Then \(1\le \lambda ^{-1}\le 6\) and
\(\square \)
We remark that the duality trick which we have used in the third case is only needed to get a better bound when \(\lambda \) is close to zero. We could indeed modify the proof in the second case in order to get a result, depending on a fixed \({\varepsilon }\in (0,1]\), which holds for \({\varepsilon }\le \lambda \le 1\).
As promised, we now study the special algebraic numbers \(t_0\) such that
Then \(t_0\) trivially satisfies the equation \(\alpha t^\lambda +\beta (1-t)^\lambda =~1\).
Let us first assume \(\lambda =1\). Then (A.2) is satisfied if and only if \(\alpha =\beta =1\) and in this case the equation \(\alpha t^\lambda +\beta (1-t)^\lambda =1\) becomes trivial. Thus we do not have bounded height.
Let now suppose \(\lambda \ne 1\). Since
we still have bounded height, but the bound seems to go to infinity as \(\lambda \rightarrow 1\), unless \(\alpha \), \(\beta \) are both roots of unity. However we have the additional equation
and it may be seen in several way that if \(\alpha \), \(\beta \) are not both roots of unity, this equation determines at most finitely many values of \(\lambda \ne 1\) in terms of \(\alpha \) and \(\beta \).
For example, this is one of the few effective instances of the Skolem-Mahler-Lech Theorem, which follows from linear forms in logarithms combining [12, Theorem 1, p. 65] with Kummer theory. Or we could simply apply Liardet’s Theorem (as made effective for example by Bérczes, Evertse and Györy [3, Theorem 2.3]); this possibility was mentioned in [6, pp. 1121–1122].
In any case we get an effective bound \(h(t_0)\le C(\alpha ,\beta )\).
Examples of (A.4) are
for positive integer q with \(t_0=2^{-q}\). Letting \(q\rightarrow \infty \) shows that such a \(C(\alpha ,\beta )\) cannot be bounded as any function of \(h(\alpha )+h(\beta )\).
In the special case \(\alpha =\beta =1\), the solutions \(t_0\) of the Eq. (A.3) are roots of unity, and thus, by the previous remarks, any solution of \(t^\lambda +(1-t)^\lambda =1\) with \(\lambda >0\) and \(\lambda \ne 1\) satisfies \(h(t_0)\le 100\max (1,\lambda ^{-1})\). This completes the proof of Theorem 1.2 stated in the introduction.
Appendix B: Exceptional solutions of the Denz equation
We return to the system (6.1); this will finally settle the Denz equation as in Theorem 1.4.
Proposition B.1
Suppose \(\delta \) is rational with \(0<|\delta |\le 10^{-330}\). Then there is no \(t_0 \ne 0,1,-1\) for which the determinations satisfy
We will need the following “two-circle” results, in which the adjectives refer to their mode of intersection.
Lemma B.2
For complex z, w with \(|z|=|w|=1\) we have
Proof
With \(z=e^{i\theta },w=e^{i\phi }\) this is equivalent (after squaring) to \(F \ge 0 ~(0 \le \theta ,\phi \le 2\pi )\) with
We have
which vanishes in the interior only for \(\theta \equiv \pi +2\phi \) modulo \(2\pi {\mathbb {Z}}\). At these points \(F=G(\cos \phi )\) with
for \(-1 \le x \le 1\). And on the boundary \(F(\theta ,0)=F(\theta ,2\pi )=H(\cos \theta )\) with
and \(F(0,\phi )=4\). This completes the proof. \(\square \)
The constant 3 here is best possible, as the example \(z=-1,w=1\) shows.
Lemma B.3
For complex z, w with \(|z|=|w|=1\) we have
Proof
With \(\epsilon =|z+2w-1|\) we have
so \(|z-1| \ge 2-\epsilon \). If \(\epsilon \le 2\) then drawing a picture shows that
And if \(\epsilon >2\) then even
which completes the proof. \(\square \)
The exponent 1/2 comes from tangency. The multiplying constant 2 cannot be replaced by anything smaller, as the example
with
and \(\epsilon \rightarrow 0\) shows.
Proof of Proposition B.1
To keep better track of the determinations and their Galois conjugates, we write \(\delta =r/q\) for integers \(q>0,r \ne 0\) and consider the set \(Y_{rq}\) in affine \(\textbf{A}^4\) defined by the equations
(and \(y \ne 0,1,-1\)). Note that this is a finite set in \({\overline{\mathbb {Q}}}^4\).
One checks that there is an involution on \(Y_{rq}\) defined by sending \((y,y_1,y_2,y_3)\) to
Now a solution of (B.1) leads to a point
on \(Y_{rq}\). So it suffices to show that \(Y_{rq}\) is empty whenever
Any element \(\sigma \) of \(\textrm{Gal}({\overline{\mathbb {Q}}}/{\mathbb {Q}})\) acts on points \((\eta ,\eta _1,\eta _2,\eta _3)\) of \(Y_{rq}\).
Case 0. For some \(\sigma \) we have
We note that for any z with \(1/2 \le |z| \le 3\) we have
Thus with \(z=\sigma (\eta ),1-\sigma (\eta ),1+\sigma (\eta )\) and corresponding \(s_1=\sigma (\eta _1),s_2=\sigma (\eta _2),s_3=\sigma (\eta _3)\), taking conjugates in (B.2) gives
and
(note that \(s_1^q=\sigma (\eta )^r\) so \(|s_1|=|\sigma (\eta )|^\delta \), and similarly for \(|s_2|,|s_3|\)). As
we get for \(s_1'=s_1/|s_1|,s_2'=s_2/|s_2|,s_3'=s_3/|s_3|\) on the unit circle the inequalities
(note that \(s_1'-2s_2'+1=(s_1'-s_1)-2(s_2'-s_2)\) and so on). By Lemma B.3 with \(z=-s_1',w=s_2'\) we get
and then with \(z=s_1',w=s_3'\)
These contradict each other provided \(|\delta |<1/24\), certainly guaranteed by (B.4).
Thus for every \(\sigma \) there are four possibilities coming from the failure of Case 0, and we consider each in turn.
Case 1a. For some \(\sigma \) we have
We write \(\sigma (\eta )=1-u\) so that \(|u| < 1/2\). Now
so that \(\sigma (\eta _1)\) is a determination of \((1-u)^\delta \). There is a canonical such determination \(b_1=1-\delta v\) for
and because
we deduce
Thus \(\sigma (\eta _1)=b_1\zeta _1\) for a root of unity \(\zeta _1\).
(Almost) similarly \(\sigma (\eta _3)^{2q}=(2-u)^{2r}\) and so there is a root of unity \(\zeta \) such that \(\zeta \sigma (\eta _3)\) is a determination of \((2-u)^\delta \), and there is a canonical \(b_3\) with
(with of course the natural determination of \(2^\delta \)). So \(\sigma (\eta _3)=b_3\zeta _3\) for a root of unity \(\zeta _3\).
The second equation in (B.2) now leads to
so that by (B.5) and (B.6) we get first \(|\zeta _1+2^{1+\delta }\zeta _3-1| \le (2^{1+\delta }+1)|\delta |\) and then
Now Lemma B.3 gives
The first equation in (B.2) gives \(2\sigma (\eta _2)=1+b_1\zeta _1=(b_1-1)\zeta _1+(\zeta _1+1)\) so we deduce
by (B.5) and (B.9). Similarly \(\sigma (\eta _1)+1=1+b_1\zeta _1\) leads to
We may say that these \(\sigma (\eta _1)\) cluster near \(-1\). Also
Now \(\eta _2^{2q}=(1-\eta )^{2r}\) gives \(|\sigma (\eta _2)|=|1-\sigma (\eta )|^\delta \). Thus
by(B.10) and the basic assumption in this Case 1a.
We leave this case unresolved for the moment, and proceed straight to
Case 1b. For some \(\sigma \) we have
Here the same arguments with \(-b_1\zeta _1+2b_2\zeta _2=1\) instead of (B.7) lead to
in place of (B.11), so clustering at 1, but also
as in (B.12), as well as
as in (B.13). Again we jump to the next (somewhat critical)
Case 2a. For some \(\sigma \) we have
Now subtracting the two equations in (B.2) and using \(\sigma (\eta _2)=b_2\zeta _2\) with \(|b_2-1| \le |\delta |\) and \(\sigma (\eta _3)=b_3\zeta _3\) with \(|b_3-1| \le |\delta |\) leads to
as in (B.8). By Lemma B.2 we deduce \(|\zeta _2^2-\zeta _2+1| \le 6|\delta |\) and then using \(|\sigma (\eta _2)-\zeta _2| \le |\delta |\) also \(|\sigma (\eta _2)^2-\sigma (\eta _2)+1| \le 12|\delta |\) and, multiplying by 4,
Here we may say that these \(\sigma (\eta _1)\) cluster near \(\pm \sqrt{-3}\).
In particular
and also
Now we get
Next we claim that \(\eta _1\) is a unit. Write \(p=-r>0\) so that \(\delta =-p/q\). Pick any \(\tau \) with \(\tau ^p=\eta _1^{-1}\). Then \(\tau ^{pq}=\eta _1^{-q}=\eta ^p\) so \(\eta =\zeta \tau ^q\) for a root of unity \(\zeta \). Now \(2\eta _2=\eta _1+1\) so
or as a polynomial in \(\tau \)
Comparing highest and lowest coefficients we see that \(\tau \) is a unit and therefore also \(\eta _1=\tau ^{-p}\) as claimed (here it is crucial that \(\delta <0\) and it is highly unlikely that it works when \(\delta >0\)).
Before proceeding further we jump to the final
Case 2b. For some \(\sigma \) we have
This is \(|\sigma (\eta ^{-1})| < 1/2\), and according to (B.3) the point \((\eta ^{-1},\eta _1^{-1},\eta _2\eta _1^{-1},-\eta _3\eta _1^{-1})\) lies in \(Y_{rq}\). By Case 2a we get
in place of (B.17), so clustering near \(\pm 1/\sqrt{-3}\), as well as
as in (B.18), and also
as in (B.19). Then
as in (B.20). Also \(\eta _1^{-1}\), so \(\eta _1\) too, is a unit.
Now let us sum up. As mentioned, we can ignore Case 0, so that every \(\sigma \) falls into the other cases.
Suppose first that \(\delta <0\).
Then Cases 1a and 1b are impossible by (B.13) and (B.16). We consider the number
which is an algebraic integer.
In Case 2a we use (B.17) and (B.19) to see that
And in Case 2b we use (B.21) and (B.22) to get the same inequality.
Thus we see that \(\eta '\) has norm of absolute value at most \((192|\delta |)^{d'}\) for its degree \(d'\). So as soon as \(|\delta | < {1 / 192}\) we deduce \(\eta '=0\).
But this would imply that \(\eta _1=i\sqrt{3}\) (say), so \(\eta _2=(1+\eta _1)/2\) and \(\eta _3=(1-\eta _1)/2\) are roots of unity so also \(1-\eta ,1+\eta \) which is impossible as \(\eta \ne 0\); and a similar argument works with say \(\eta _1=1/(i\sqrt{3})\) using \(\eta ^{-1}\ne 0\). This settles things when \(\delta <0\).
It remains to deal with \(\delta >0\).
Then Cases 2a and 2b are impossible by (B.20) and (B.23). But now we no longer know that \(\eta _1\) is a unit (and probably it need not be), so we cannot use this method.
Instead we use Theorem A.1 on the equation \(\eta _1-2\eta _2=-1\). We note that \(\eta _1^q=\eta ^r\) so \(\eta _1\) is a determination of \(\eta ^\delta \). Also \(\eta _2^{2q}=(1-\eta )^{2r}\) so there is a root of unity \(\zeta \) such that \(\eta _2/\zeta \) is a determination of \((1-\eta )^\delta \). Now
for \(\alpha =-1,\beta =2\zeta \). It follows from Theorem A.1 with \(\lambda =\delta >0\) that
(as long as we don’t have \(\alpha \eta ^{\delta -1} = 1\), in which case \(h(\eta )=0\) anyway). Therefore
Now we consider \(\eta '=\eta _1^2-1\), so that
In Case 1a we use (B.11) and (B.12) to get
And in Case 1b the same using (B.14) and (B.15).
If \(\eta ' \ne 0\) the Product Formula gives \(1 < (30|\delta |^{1/2})^{d'}e^{d'h(\eta ')}\) which by (B.24) is at most \((30|\delta |^{1/2}e^{200}2^{243})^{d'}\). So as soon as
(accounting for (B.4) above) we conclude \(\eta '=0\), now easily seen to be impossible. This settles things when \(\delta >0\), thereby completing the proof of the Proposition. \(\square \)
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Amoroso, F., Masser, D. & Zannier, U. Bounded height in pencils of subgroups of finite rank. Math. Ann. (2023). https://doi.org/10.1007/s00208-023-02724-5
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00208-023-02724-5