Number of integral points on quadratic twists of elliptic curves

Basic Information

• Paper ID: 2509.03274
• Title: Number of integral points on quadratic twists of elliptic curves
• Author: Seokhyun Choi (KAIST)
• Classification: math.NT (Number Theory)
• Publication Date: November 12, 2025
• Paper Link: https://arxiv.org/abs/2509.03274v2

Abstract

This paper investigates the integral point problem on quadratic twists $E_D : y^2 = x^3+D^2Ax+D^3B$ of elliptic curves $E : y^2 = x^3+Ax+B$ over the rationals $\mathbb{Q}$. For sufficiently large $D$, the author proves that the number of integral points on $E_D$ satisfies the upper bound $\ll 4^r$, where $r$ denotes the Mordell-Weil rank of $E_D$, and the implicit constant is absolute (independent of the elliptic curve $E$).

Research Background and Motivation

Problem Background

1. Siegel's Theorem (1929): The set of integral points $E(\mathbb{Z})$ on an elliptic curve is finite, but the result is not effectively computable due to its dependence on the Thue-Siegel-Roth theorem.
2. Development of Effective Bounds:
   • Baker (1966): First effective bound on the height of integral points, but extremely large: $|x| \leq \exp((10^6\max\{|A|,|B|\})^{10^6})$
   • Lang's Conjecture: There exists an absolute constant $C$ such that $|E(\mathbb{Z})| \ll C^r$
3. Existing Results:
   • Silverman (1987): Proved Lang's conjecture for elliptic curves with bounded Szpiro ratio
   • Helfgott-Venkatesh (2006): Proved $|E(\mathbb{Z})| \ll C^{\omega(\Delta)}(\log|\Delta|)^2(1.34)^r$
   • For quadratic twist families: Gross-Silverman gave $C \sim 10^9$, Chi-Lai-Tan improved to $C=25$, Chan improved to $C=3.8$ for congruent number curves

Research Motivation

This paper establishes an improved constant bound $C=4$ for the quadratic twist family, representing significant progress on this problem. The result holds for sufficiently large $D$, with an absolute and effectively computable constant.

Core Contributions

1. Main Theorem (Theorem 1.1): For quadratic twists $E_D$ of an elliptic curve $E : y^2 = x^3+Ax+B$, when $D$ is sufficiently large, it is proved that $|E_D(\mathbb{Z})| \ll 4^r$, where $r$ is the rank of $E_D$.
2. Methodological Innovation: Combines gap principles and Diophantine approximation techniques, partitioning the integral point set into four categories (small, medium-small, medium-large, and large points) for separate treatment.
3. Technical Improvements:
   • Applies spherical codes theory for small and medium-sized points
   • Uses quantitative Roth's theorem for large points
   • Employs refined height estimates and geometric analysis
4. Constant Optimization: Improves the constant for quadratic twist families from 25 to 4, significantly enhancing the precision of the bound.

Detailed Methodology

Overall Strategy

The author partitions $E_D(\mathbb{Z})$ into four subsets based on the canonical height $\hat{h}(P)$:

1. Small points: $E_D(\mathbb{Z})_{\text{small}} := \{P \in E_D(\mathbb{Z}) \mid \hat{h}(P) \leq 1.5\log D\}$
2. Medium-small points: $E_D(\mathbb{Z})_{\text{medium-small}} := \{P \in E_D(\mathbb{Z}) \mid 1.5\log D \leq \hat{h}(P) \leq 20\log D\}$
3. Medium-large points: $E_D(\mathbb{Z})_{\text{medium-large}} := \{P \in E_D(\mathbb{Z}) \mid 20\log D \leq \hat{h}(P) \leq 2^{200}\log D\}$
4. Large points: $E_D(\mathbb{Z})_{\text{large}} := \{P \in E_D(\mathbb{Z}) \mid \hat{h}(P) \geq 2^{200}\log D\}$

Technical Tools

1. Height Estimates (Lemmas 3.1-3.4)

Establishes relationships between canonical height $\hat{h}$ and Weil height $h$:

• For $P \in E_D(\mathbb{Q})$: $c_1 \leq \hat{h}(P) - h(P) \leq c_2$
• For $P \in E_D(\mathbb{Q})$ (via isomorphism $\phi_D: E_D \to \bar{E}_D$):
  • General case: $c_1 - \log D \leq \hat{h}(P) - h(P) \leq c_2 + \log D$
  • When $x(P) > D$: $c_1 - \log D \leq \hat{h}(P) - h(P) \leq c_2$

2. Gap Principle (Gap Principle, Lemma 3.9)

For integral points $P, Q \in E_D(\mathbb{Z})$ satisfying $MD \leq x(P) < x(Q)$: $h(P+Q) \leq h(P) + 2h(Q) + 2.9$

This inequality is central to subsequent analysis, controlling the "distance" between points.

3. Spherical Codes Theory (Section 4)

On $E_D(\mathbb{Q}) \otimes_{\mathbb{Z}} \mathbb{R} \cong \mathbb{R}^r$, the canonical height induces an inner product structure. Define the angle between two points $P, Q$: $\cos\theta_{P,Q} := \frac{\hat{h}(P+Q) - \hat{h}(P) - \hat{h}(Q)}{2\sqrt{\hat{h}(P)\hat{h}(Q)}}$

If a collection of points satisfies $\cos\theta_{P,Q} \leq \cos\theta_0$, they form a spherical code with cardinality at most $A(r, \theta_0)$.

Two bounds are employed:

• Kabatiansky-Levenshtein bound ($0 < \theta < \pi/2$): $A(r,\theta) \ll \left[\exp\left(\frac{1+\sin\theta}{2\sin\theta}\log\frac{1+\sin\theta}{2\sin\theta} - \frac{1-\sin\theta}{2\sin\theta}\log\frac{1-\sin\theta}{2\sin\theta} + 0.001\right)\right]^r$
• Simple bound ($\theta > \pi/2$): $A(r,\theta) \ll 1$

Treatment of Different Point Classes

Small Points (Proposition 5.1)

• Decomposes $E_D(\mathbb{Z})_{\text{small}}$ into cosets of $4E_D(\mathbb{Q})$
• For each coset $S(R)$, proves that any two distinct points $P_1, P_2 \in S(R)$ satisfy: $\cos\theta_{P_1,P_2} \leq -\frac{1}{6} < 0$
• By the spherical code bound for $\theta > \pi/2$, $|S(R)| \ll 1$
• With at most $4^{r+1}$ cosets, $|E_D(\mathbb{Z})_{\text{small}}| \ll 4^r$

Medium-Small Points (Proposition 6.3)

• Further subdivides into $MS_n := \{P \mid (n-0.5)\log D \leq \hat{h}(P) \leq (n+0.5)\log D\}$, $2 \leq n \leq 20$
• Only considers points with $y(P) > 0$ (denoted $MS_n^+$)
• Uses Lemma 6.1: when $x(P) \leq MD$, $\hat{h}(P) < 1.5\log D$ (excluded)
• For $P, Q \in MS_n^+$, applies gap principle and optimized analysis from Lemma 6.2: $\cos\theta_{P,Q} \leq \max\left\{\frac{n+1.6}{2\sqrt{n^2-0.25}}, 1-\frac{n-1.6}{2(n+0.5)}\right\}$
• Applying Kabatiansky-Levenshtein bound yields $|MS_n^+| \ll 4^r$

Medium-Large Points (Proposition 7.2)

• Subdivides into $ML_n := \{P \mid 20 \cdot (1.1)^{n-1}\log D \leq \hat{h}(P) \leq 20 \cdot (1.1)^n\log D\}$, $1 \leq n \leq 50$
• For $P, Q \in ML_n^+$ satisfying $\max\{\hat{h}(Q)/\hat{h}(P), \hat{h}(P)/\hat{h}(Q)\} \leq 1.1$
• Lemma 7.1 proves: $\cos\theta_{P,Q} \leq 0.63$
• Spherical code theory yields $|ML_n^+| \ll (1.55)^r \ll 4^r$

Large Points (Proposition 10.4)

This is the most complex case, handled in two steps:

Step 1: Decomposes into cosets of $3E_D(\mathbb{Q})$, denoted $L(R)$, further subdivided into:

• $L(R)^* := \{P \in L(R) \mid \hat{h}(P) \leq 10^{50}\hat{h}(R)\}$
• $L(R)^{**} := \{P \in L(R) \mid \hat{h}(P) > 10^{50}\hat{h}(R)\}$

For $L(R)^*$, similar to medium-large points, subdivides into 700 subintervals (ratio 1.01), using Lemma 10.1 to prove $\cos\theta_{P,Q} \leq 0.504$, yielding $|L(R)^*| \ll (1.33)^r$.

Step 2: For $L(R)^{**}$, uses Diophantine approximation:

• Writes $P = 3Q + R$, choosing $S \in \frac{1}{3}R$ to minimize $|x(Q) - x(S)|$
• Key Lemma (Lemma 8.2): When $h(P) > 1000h(R)$ and $h(P) > 2000\log D$: $\frac{\log|x(Q) - x(S)|}{h(Q)} < -2.75$ This means $x(Q)$ is a rational approximation to the algebraic number $x(S)$ with approximation exponent $> 2.75 > 2$
• Applies Quantitative Roth's Theorem (Evertse, Theorem 9.1): For an algebraic number $\alpha$ of degree $\leq 9$, the number of rational numbers $\beta$ with approximation exponent $> 2.75$ is at most: $225 \cdot 0.75^{-3} \log(18) \log(0.75^{-1}\log(18)) \ll 1$
• Therefore $|L(R)^{**}| \ll 1$

Combining both steps: $|L(R)| \ll (1.33)^r$, with at most $3^r$ cosets, so $|E_D(\mathbb{Z})_{\text{large}}| \ll (3.99)^r \leq 4^r$.

Technical Innovations

1. Refined Interval Partitioning: Through logarithmic-scale fine partitioning (1.5, 20, $2^{200}$ and internal ratios 1.1, 1.01), establishes tighter gap principles within each interval.
2. Geometric Analysis: Utilizes geometric relationships of $x$-coordinates (Lemmas 3.6-3.8), precisely estimating coordinate changes after point addition, crucial for establishing gap principles.
3. Optimization Techniques: Lemma 6.2 provides analysis of maximum values of binary functions over rectangular regions, optimizing angle bound calculations.
4. Refined Application of Diophantine Approximation: Through construction of trisection points and discriminant estimates (Lemma 8.1), transforms the large point problem into an effective Diophantine approximation problem.

Experimental Setup

This is a pure mathematics theory paper with no numerical experiments. All results are rigorous mathematical proofs.

Computational Verification

Appendix B provides numerical tables of $\cos\theta$ and spherical code bounds $E(\theta)$ for various $n$ values in Proposition 6.3, for example:

• $n=2$: $\cos\theta = 0.9295$, $E(\theta) = 3.603$
• $n=10$: $\cos\theta = 0.6000$, $E(\theta) = 1.482$
• $n=20$: $\cos\theta = 0.5512$, $E(\theta) = 1.397$

These numerical values verify the effectiveness of the theoretical bound $\ll 4^r$.

Experimental Results

Main Results

Theorem 1.1: For the quadratic twist $E_D: y^2 = x^3 + D^2Ax + D^3B$, when $D$ is sufficiently large: $|E_D(\mathbb{Z})| \ll 4^r$ where the implicit constant is absolute (independent of $A, B$) and effectively computable.

Comparison with Existing Results

This paper improves the constant for general quadratic twist families to nearly match Chan's result for special curves, representing significant progress.

Verification of Technical Details

The proofs of various propositions provide specific constants:

• Proposition 5.1: $|E_D(\mathbb{Z})_{\text{small}}| \ll 4^r$
• Proposition 6.3: $|E_D(\mathbb{Z})_{\text{medium-small}}| \ll 4^r$
• Proposition 7.2: $|E_D(\mathbb{Z})_{\text{medium-large}}| \ll (1.55)^r \ll 4^r$
• Proposition 10.4: $|E_D(\mathbb{Z})_{\text{large}}| \ll (3.99)^r \leq 4^r$

The final constant 4 is primarily determined by the large point analysis: $(1.33)^r \times 3^r = (3.99)^r$.

Related Work

Historical Development

1. Classical Theory:
   • Siegel (1929): Finiteness of integral points (not effectively computable)
   • Baker (1966): First effective bound (very large)
   • Lang (1970s): Proposes $|E(\mathbb{Z})| \ll C^r$ conjecture
2. General Elliptic Curves:
   • Silverman (1987): Bounded Szpiro ratio case
   • Hindry-Silverman (1988): Canonical height method
   • Helfgott-Venkatesh (2006): $C = 1.34$ (with logarithmic factors)
3. Quadratic Twist Families:
   • Silverman (1987): Establishes basic framework
   • Gross-Silverman (1995): $C \sim 10^9$
   • Helfgott (2004): Introduces gap principle
   • Chi-Lai-Tan (2005): $C = 25$
   • Alpoge (2014): Average bounds research
   • Chan (2022): Congruent number curves $C = 3.8$

Innovations in This Paper

Compared to previous work:

• Methodology: Combines gap principles, spherical codes theory, and quantitative Roth's theorem into a systematic framework
• Technical Refinement: Through multi-level interval partitioning and precise geometric estimates, significantly improves the constant
• Universality: Applies to all quadratic twists, not limited to special curves

Conclusions and Discussion

Main Conclusions

This paper proves that for quadratic twists $E_D$ of elliptic curves, when $D$ is sufficiently large, the number of integral points satisfies $|E_D(\mathbb{Z})| \ll 4^r$, where the constant 4 is absolute and effectively computable. This represents one of the best-known results for Lang's conjecture on quadratic twist families.

Limitations

1. "Sufficiently Large" Condition: The theorem requires $D \geq D_0$, where $D_0$ depends on $A, B$. Although theoretically $D_0$ is computable, specific values are not provided in the paper.
2. Absoluteness of Constants: While the constant 4 is independent of $E$, the multiplicative constant hidden in $\ll$ may be large, requiring more precise estimates for practical applications.
3. General Elliptic Curves: This paper only treats quadratic twist families. For general elliptic curves (non-twist cases), the optimal constant in Lang's conjecture remains unknown.
4. Small $D$ Cases: The paper's methods do not apply to bounded $D$, requiring alternative techniques (such as Baker's method).

Future Directions

1. Further Constant Improvement: Can the constant 4 be reduced to approach the theoretical optimum (possibly near 3 or smaller)?
2. Effectiveness: Provide explicit bounds for $D_0$, making results applicable in practical computations.
3. Generalization to Other Twists: Extend the methods to cubic twists, quartic twists, and more general twist families.
4. Connection to BSD Conjecture: Explore relationships between the number of integral points and arithmetic invariants such as L-functions and Tate-Shafarevich groups.
5. Average Bounds: Study average behavior of integral point counts over all quadratic twists (direction of Alpoge's work).

In-Depth Evaluation

Strengths

1. Important Theoretical Progress: Improving the constant for quadratic twist families from 25 to 4 represents a significant breakthrough in the field, approaching the special result 3.8 for congruent number curves.
2. Systematic Methodology:
   • Four-layer decomposition strategy is clear and reasonable
   • Each layer employs the most suitable tools (spherical codes, Diophantine approximation)
   • Technical combination is elegant and efficient
3. Technical Depth:
   • Refined application of gap principles (Lemma 3.9 and corollaries)
   • Precision of geometric analysis (coordinate estimates in Lemmas 3.6-3.8)
   • Clever construction in Diophantine approximation (trisection point technique in Lemma 8.2)
4. Rigor of Proofs: All estimates include detailed calculations with complete verification of key inequalities in appendices (Appendix A, B).
5. Clear Exposition:
   • Well-organized hierarchical structure
   • Systematic lemma numbering
   • Sufficient explanation of key ideas

Weaknesses

1. Computability Limitations:
   • Specific bound for "sufficiently large $D$" not explicitly stated
   • Magnitude of implicit constants not discussed
   • Limited practical applicability
2. Unknown Tightness of Constants:
   • Does there exist $E_D$ of rank $r$ with $|E_D(\mathbb{Z})| \sim 4^r$?
   • What is the theoretically optimal constant?
   • How does it compare to Helfgott-Venshtein's 1.34 (despite logarithmic factors)?
3. Technical Limitations:
   • Methods highly dependent on special structure of quadratic twists
   • Difficult to generalize to general elliptic curves
   • Ineffective for small $D$
4. Absence of Numerical Verification: Although a theory paper, lacks concrete examples verifying the effectiveness of theoretical bounds.

Impact

1. Theoretical Significance:
   • Advances research on Lang's conjecture
   • Provides new technical tools for integral point problems
   • May inspire research on other Diophantine equations
2. Methodological Contribution:
   • Systematic application of gap principles + spherical codes
   • Multi-scale decomposition strategy
   • Effective use of quantitative Roth's theorem
3. Future Research:
   • May stimulate further constant optimization
   • Provides methodological template for other twist families
   • Connects to other problems in arithmetic geometry
4. Reproducibility: Proof is entirely theoretical with strong reproducibility, though numerical implementation requires additional work.

Applicable Scenarios

1. Theoretical Research:
   • Arithmetic theory of elliptic curves
   • Diophantine geometry
   • Height theory
2. Potential Applications:
   • Elliptic curve parameter selection in cryptography
   • Congruent number problem
   • Theoretical foundation for rational point search algorithms
3. Educational Value: Demonstrates comprehensive application of multiple modern number-theoretic techniques, suitable as case study for advanced courses.

Key References

1. 21 Siegel (1929): Foundational work on finiteness of integral points
2. 16,17 Lang (1970s-1980s): Proposal of Lang's conjecture
3. 22 Silverman (1987): Quantitative Siegel theorem, establishing height methods
4. 12 Helfgott (2004): Introduction of gap principle
5. 14 Helfgott-Venkatesh (2006): Result with $C=1.34$
6. 5 Chi-Lai-Tan (2005): Previous optimal result with $C=25$
7. 4 Chan (2022): Congruent number curves with $C=3.8$
8. 9 Evertse (2010): Quantitative Roth's theorem
9. 15 Kabatiansky-Levenshtein (1978): Spherical codes bound

Overall Assessment: This is a high-quality number theory paper achieving substantial progress on Lang's conjecture. The methodology is systematic, proofs are rigorous, and results are significant. While there remains room for improvement in computability and constant tightness, the paper represents the current state-of-the-art for this problem. It possesses important theoretical value and methodological significance for research on integral points on elliptic curves.