Small gaps between Goldbach primes

Basic Information

• Paper ID: 2508.02769
• Title: Small gaps between Goldbach primes
• Author: Mizuki Akeno (Faculty of Mathematics, University of Tsukuba)
• Classification: math.NT (Number Theory)
• Publication Date: November 25, 2025 (arXiv v2)
• Paper Link: https://arxiv.org/abs/2508.02769v2

Abstract

This paper investigates the problem of small gaps in the set of Goldbach primes $\mathcal{P}(N) = \mathbb{P} \cap (N - \mathbb{P})$ using the Bombieri-Davenport method and the Maynard-Tao method, with comparative analysis of both approaches. The main results include: (1) Using the Bombieri-Davenport method, we prove that for almost all even integers $N$, the minimal gap in $\mathcal{P}(N)$ is at most $0.765\ldots$ times the average gap, improving Tsuda's result; (2) We prove that direct application of the Maynard-Tao method is insufficient to improve this bound; (3) However, the Maynard-Tao method can establish bounded gaps between Goldbach primes with bounded error for almost all even integers $N$.

Research Background and Motivation

Research Problem

The core problem studied in this paper is: for an even integer $N$, define the set of Goldbach primes as $\mathcal{P}(N) = \{p \leq N : \exists p' \in \mathbb{P}, p + p' = N\} = \mathbb{P} \cap (N - \mathbb{P})$ and investigate how small the minimal gaps between primes in this set can be.

Importance of the Problem

1. Refinement of the Goldbach Conjecture: While the classical Goldbach conjecture concerns whether even integers can be expressed as sums of two primes, this paper further investigates the distribution density of primes satisfying this property.
2. New Perspective on Prime Gaps: This is a study of the prime gap problem under special constraints, connecting two important problems in number theory.
3. Development of Sieve Methods: By comparing two important sieve methods (Bombieri-Davenport and Maynard-Tao), we deepen our understanding of the capabilities and limitations of sieve methods.

Limitations of Existing Methods

1. Mikawa-Nakamura (unpublished): Proved $\Xi \leq 5/6 = 0.8333\ldots$
2. Tsuda (2024): Proved $\Xi \leq 0.8201\ldots$
3. Here $\Xi$ is defined as: for any $\epsilon, \varepsilon > 0$, almost all even integers $N$ satisfy $\min_{p,p' \in \mathcal{P}(N), N^{1-\epsilon} < p' < p} (p - p') \leq (\Xi + \varepsilon)S(N)^{-1}(\ln N)^2$ where $S(N)$ is the singular series.

Research Motivation

1. Improve Known Bounds: Improve the distribution level through simple observations, enhancing the results of the Bombieri-Davenport method.
2. Method Comparison: Systematically compare the merits and drawbacks of both methods, revealing the limitations of the Maynard-Tao method on this problem.
3. Establish Bounded Gaps: Although we cannot improve the bound on $\Xi$, we prove the existence of absolutely bounded gaps.

Core Contributions

1. Improved Upper Bound for $\Xi$: We prove $\Xi \leq \frac{7}{72}(4 + \sqrt{15}) = 0.76542\ldots$, improving Tsuda's result of 0.8201 (Theorem 1.2).
2. Reveal Limitations of Maynard-Tao Method: We prove that direct application of the Maynard-Tao method cannot improve the above bound, and is limited by $\Xi \leq 0.826\ldots$ (Section 5.1).
3. Establish Existence of Bounded Gaps: We prove that there exists $H = 56250000 = (7500)^2$ such that for almost all even integers $N$, there exist prime pairs in $\mathcal{P}_H(N)$ with gaps not exceeding $H$ (Theorem 1.3).
4. Improve Distribution Level: By observing that we can reverse the roles of $f_1, f_2$, we extend the distribution level from $[0,1/2] \times [0,1/3]$ to $([0,1/2] \times [0,1/3]) \cup ([0,1/3] \times [0,1/2])$ (improved Theorem 2.1).
5. Methodological Contribution: We provide a method for evaluating weighted sums different from those of Maynard and Tao, avoiding difficulties arising from the W-trick.

Detailed Methods

Task Definition

Main Constant Definitions:

• $\Xi$: The infimum of the ratio of minimal gap to average gap (relative version)
• $\Xi^*$: Upper bound for minimal gap in absolute sense (if it exists)

Objectives:

1. Provide the smallest possible upper bound for $\Xi$
2. If possible, prove that $\Xi^*$ exists and provide an upper bound

Core Technical Framework

This paper employs two main methods:

1. Bombieri-Davenport Method

Basic Idea: Analyze the second moment $M_2 = \sum_{m+n=N} \left(\sum_{h \leq H} \Lambda(m+h)\Lambda(n-h)\right)^2$ where $H = cS(N)^{-1}(\ln N)^2$.

Key Inequality: Use the trivial inequality $0 \leq \sum_{m+n=N} \left(\sum_{h \leq H} \Lambda(m+h)\Lambda(n-h) - \sum_{h \leq H} (\Lambda(m+h)\Lambda(n-h))^\sharp\right)^2$

where $(\Lambda(m)\Lambda(n))^\sharp$ is an appropriate approximation to $\Lambda(m)\Lambda(n)$: $(\Lambda(m)\Lambda(n))^\sharp = \Lambda_{X_2}(m)\Lambda_{X_3}(n) + \Lambda_{X_3}(m)\Lambda_{X_2}(n) - \Lambda_{X_3}(m)\Lambda_{X_3}(n)$ with $X_r = X^{(1-\varepsilon)/r}$.

Innovative Improvement:

• Utilize the symmetry of the distribution level, extending the region $D$ from $[0,1/2] \times [0,1/3]$ to $D = ([0,1/2] \times [0,1/3]) \cup ([0,1/3] \times [0,1/2])$
• This gives $\text{meas}(D) = \frac{7}{9}$, thereby improving the bound.

Upper Bound Sieve Combination: Use the Rosser-Iwaniec upper bound sieve (Lemma 4.3): $\sum_{m+n=N, N^{1-\epsilon} < m,n} \Omega(m+h)\Lambda(n-h)\Lambda(m+h')\Lambda(n-h') \leq 16NS(\{h,h'\}; N)(1+\varepsilon)$

By choosing $H_+ = \kappa S(N)^{-1}(\ln N)^2$ and $H_- = \lambda S(N)^{-1}(\ln N)^2$, we optimize to obtain $\lambda > \frac{7}{72}(4 + \sqrt{15})$ when the inequality holds.

2. Maynard-Tao Method

Weight Selection: Consider weights $w(m,n;\mathcal{H}) = \left(\sum_{\substack{d_i | n+h_i, 1 \leq i \leq k \\ e_i | m-h_i, 1 \leq i \leq k}} \lambda_{d_1,\ldots,d_k,e_1,\ldots,e_k}\right)^2$

adopting Tao's choice: $\lambda_{d_1,\ldots,d_{2k}} = \mu(d_1)\cdots\mu(d_{2k})F\left(\frac{\ln d_1}{\ln X}, \ldots, \frac{\ln d_{2k}}{\ln X}\right)$

Key Lemma 3.2: Establish an asymptotic formula; for almost all $N \in (X/2, X]$: $\sum_{m+n=N} \prod_{i=1}^k \Lambda_{Q_i}(m+g_i)\Lambda_{R_i}(n+h_i) = N(S(\mathcal{G}, \mathcal{H}; N) + o(1))$ when $(\sum \alpha_i, \sum \beta_i) \in (1-\varepsilon) \cdot [0,1] \times [0,1]$.

Variational Problem: Define $M_k(D) = \sup_f \frac{\sum_{\ell=1}^k \sum_{\ell'=1}^k J_{\ell,\ell'}^{(2k)}(f)}{I_{2k}(f)}$ where integral functionals are defined on functions $f$ satisfying $R + R \subseteq D_k$.

Limitation Analysis (Proposition 5.4): Using Cauchy-Schwarz inequality, we prove $M_k(D) \leq \left(\frac{1}{2}\max_{(s,t) \in D}(s+t)\right)^2$

For $D = ([0,1/2] \times [0,1/3]) \cup ([0,1/3] \times [0,1/2])$, this gives $M_k(D) \leq \left(\frac{1}{2} \cdot \frac{5}{6}\right)^2 = \frac{25}{144} \approx 0.174$ Therefore $\Xi \leq 1 - 0.174 = 0.826\ldots$, which cannot improve the Bombieri-Davenport method's result.

Technical Innovations

1. Exploitation of Distribution Level Symmetry: First observation that we can exchange the roles of $f_1, f_2$, extending the usable distribution region (improved Theorem 2.1).
2. Avoidance of W-trick: Develop a method for evaluating weighted sums that does not rely on the W-trick (Lemma 5.1), since the discriminant $\Delta(\mathcal{H} \cup (\mathcal{H} - N))$ may have large prime factors.
3. Strongly Admissible Tuples (Definition 5.6): Introduce a new concept requiring that $\mathcal{H} \cup (\mathcal{H} - N)$ be admissible for all even integers $N$. Construct explicit examples (Lemma 5.8): $\mathcal{H} = \{h^2 : 1 \leq h \leq H, (h, 30) = 1\}$
4. Asymptotic Formula for Singular Series Averages (Lemma 3.4): Generalize results of Gallagher and Tsuda, proving $\sum_{\substack{1 \leq h_1, \ldots, h_k \leq H(N) \\ h_i \text{ distinct}}} S(\{h_1,\ldots,h_k\} + \{0,N\}) = (H(N)S(\{0,N\}))^k(1 + O(H(N)^{\varepsilon-1}))$

Experimental Setup

Theoretical Verification Rather Than Numerical Experiments

This paper is a pure theoretical number theory paper without traditional numerical experiments, but includes the following verifications:

1. Parameter Selection Verification (Lemma 5.9):
   • Take $k = 2000$
   • Parameter choices: $c = 0.96/\ln k$, $T = 0.97/\ln k$, $\tau = 1 - k\mu$
   • Use Mathematica numerical computation to verify the existence of a piecewise continuous function $F$ such that $\frac{\sum_{\ell=1}^k \int_0^1 \cdots \int_0^1 \left(\int_0^1 F(t_1,\ldots,t_k)dt_\ell\right)^2 dt_1\ldots dt_k}{\int_0^1 \cdots \int_0^1 F(t_1,\ldots,t_k)^2 dt_1\ldots dt_k} \geq 5.00958\ldots$
2. Constant Optimization:
   • Obtain optimal $\lambda$ by solving $\kappa^2 - \frac{7}{9}\kappa > 16(\kappa - \lambda)^2$
   • Minimum is achieved at $\kappa = \kappa'$, giving $\lambda > \frac{7}{72}(4 + \sqrt{15})$

Asymptotic Property Verification

All main results hold for "almost all" $N$, with the exceptional set having size $O(X(\ln X)^{-A})$ for any $A > 0$.

Experimental Results

Main Theorem Results

Theorem 1.2 (Bombieri-Davenport Method): $\Xi \leq \frac{7}{72}(4 + \sqrt{15}) = 0.76542\ldots$

Comparison:

• Mikawa-Nakamura: $\Xi \leq 5/6 = 0.8333\ldots$
• Tsuda (2024): $\Xi \leq 0.8201\ldots$
• This paper: $\Xi \leq 0.76542\ldots$
• Improvement magnitude: Approximately 6.7% improvement over Tsuda

Theorem 1.3 (Maynard-Tao Method): There exists $H = 56250000 = (7500)^2$ such that for any $\epsilon > 0$, almost all even integers $N$ satisfy $\min_{\substack{p,p' \in \mathcal{P}_H(N) \\ N^{1-\epsilon} < p' < p}} (p - p') \leq H$

where $\mathcal{P}_H(N) = \{p \leq N : \exists p' \in \mathbb{P}, |N - p - p'| \leq H\}$.

Method Comparison Results

Bombieri-Davenport Method:

• Advantage: Provides better relative bound $\Xi \leq 0.765\ldots$
• Region used: $D = ([0,1/2] \times [0,1/3]) \cup ([0,1/3] \times [0,1/2])$, $\text{meas}(D) = 7/9$

Maynard-Tao Method:

• Limitation: Variational problem restricted by $M_k(D) \leq (5/12)^2 \approx 0.174$
• Cannot prove $\Xi < 0.826\ldots$
• Advantage: Can establish absolutely bounded gaps $\Xi^* \leq 56250000$

Theoretical Findings

1. Upper Bound for Variational Problem (Proposition 5.4): $M_k(D) \leq \left(\frac{1}{2}\max_{(s,t) \in D}(s+t)\right)^2$ This is the essential barrier of the Maynard-Tao method.
2. Symmetry of Distribution Level: After extension to $D$, the constant 16 can be improved to $144/7 = 20.57\ldots$ (Remark 4.4).
3. Bounds for Strongly Admissible Tuples (Proposition 5.11): $H(k) \gg k^{4/3}$ where $H(k)$ is the minimal length of a strongly admissible $k$-tuple.

Technical Verification

Application of Lemma 3.1: For $|h_i| \leq X^2$, almost all $N$ have $\sum_{n \leq X} \prod_{i=1}^k \Lambda_{Q_i}(n+h_i)\Lambda_{Q_i'}(n+h_i)^{\varepsilon_i} = X\left(S(\mathcal{H}) \prod_{i=1}^k (\min(\ln Q_i, \ln Q_i'))^{\varepsilon_i} + o(1)\right)$

Distribution Level Theorem 2.1: For $d_i \leq X^{(1-\varepsilon)\alpha_i}$, $(\alpha_1, \alpha_2) \in D$, almost all $N$ have $\sum_{(d_1,d_2) \in X^{(1-\varepsilon) \cdot D}} f_1(d_1)f_2(d_2)E_R(l_1(d_1), l_2(d_2), d_1, d_2; N) \ll \frac{X}{(\ln X)^B}$

Related Work

Prime Gap Problems

1. Classical Results:
   • Bombieri-Davenport (1966): $\min_{X < p' < p \leq 2X}(p - p') \leq (1/2 + \varepsilon)\ln X$
   • Zhang-Maynard-Tao-Polymath (2014): $\min_{X < p' < p \leq 2X}(p - p') \leq 246$
2. Distribution of Goldbach Problems:
   • Montgomery-Vaughan (1975): Bounds on Goldbach exceptional sets
   • Maier-Pomerance (1990): Distribution of generalized twin primes in arithmetic progressions
   • This paper's results are refinements of these works on Goldbach prime sets

Sieve Methods

1. Bombieri-Vinogradov Theorem (Lemma 2.5): Distribution level $X^{(1-\varepsilon)/2}$
2. Maynard-Tao Method:
   • Maynard (2015): Breakthrough on small prime gaps
   • Polymath (2014): Variant of Selberg sieve
   • This paper: Reveals limitations of this method under Goldbach constraints
3. Transfer Principles:
   • Green-Tao (2006), Matomäki-Maynard-Shao (2017)
   • Matomäki-Shao (2017): Almost twin prime version of Vinogradov's theorem
   • This paper's method is more direct, giving sharper bounds on $H$

Related Technical Tools

1. Exponential Sum Estimates:
   • Balog-Perelli (1985): Small arc estimates (Lemma 2.8)
   • Matomäki (2009): Bombieri-Vinogradov type exponential sums
2. Singular Series:
   • Hardy-Littlewood: Singular series for $k$-prime tuple conjecture
   • Gallagher (1976): Prime distribution in short intervals
   • This paper: Generalization to Goldbach primes (Lemma 3.4)

Unique Contributions of This Paper

Compared to related work, this paper:

1. First systematically compares two major sieve methods on the Goldbach prime gap problem
2. Reveals the variational problem barrier of the Maynard-Tao method
3. Introduces the concept of strongly admissible tuples with explicit constructions
4. Develops a technical approach that avoids the W-trick

Conclusions and Discussion

Main Conclusions

1. Improved Relative Bound: We prove $\Xi \leq 0.76542\ldots$, meaning that for almost all even integers $N$, the minimal gap in the Goldbach prime set $\mathcal{P}(N)$ does not exceed 76.5% of the average gap.
2. Method Comparison: The Bombieri-Davenport method outperforms the directly applied Maynard-Tao method on this problem, with the latter limited by the upper bound of the variational problem.
3. Absolutely Bounded Gaps: Although we cannot prove $\Xi^* < \infty$, we establish the existence of bounded gaps with bounded error ($H = 56250000$).
4. Technical Insights: The symmetry of the distribution level is the key improvement point; avoiding the W-trick makes the method more robust.

Limitations

1. Gap Between $\Xi$ Bound and Conjectured Value:
   • The generalized Hardy-Littlewood conjecture predicts $\Xi = 0$
   • Current best result $\Xi \leq 0.765\ldots$ still has substantial gap
   • Even the folklore result $\Xi \leq 1/6 = 0.1666\ldots$ (Proposition 1.4) remains unattained
2. Essential Barrier of Maynard-Tao Method:
   • Proposition 5.4 shows $M_k(D) \leq (1/2 \max_{(s,t) \in D}(s+t))^2$
   • Even assuming $D = [0,1] \times [0,1]$, cannot prove existence of $\Xi^*$
   • Requires more general setting (such as equation (53)) to break through
3. Constant for Bounded Gaps:
   • $H = 56250000$ is very large, not numerically optimized
   • No quadratic optimization used to determine sieve weights
   • No computer-aided search for narrow admissible tuples
4. Technical Limitations:
   • Distribution level restricted by Bombieri-Vinogradov theorem's $X^{1/2-\varepsilon}$
   • If Bombieri-Friedlander-Iwaniec type estimates to $X^{4/7-\varepsilon}$ could be proved, further improvement possible
   • Good factorization weight results for upper bound sieves 16, 21 may bring minor improvements

Future Directions

1. Method Improvements:
   • Problem 5.5: Can we establish existence of $\Xi^*$ under more general setting (equation (53)) with $D = [0,1] \times [0,1]$?
   • Fully exploit distribution level, develop modified versions of Maynard-Tao method
2. Constant Optimization:
   • Numerically optimize $H(k)$ (minimal length of strongly admissible $k$-tuple)
   • Problem 5.12: What is the value of $\lim_{k \to \infty} \frac{\ln H(k)}{\ln k}$?
   • Current bounds: $H(k) \gg k^{4/3}$, $H(k) \leq (k \ln k)^2(1 + o(1))$
3. Conditional Results:
   • Under Generalized Riemann Hypothesis: $\Xi' = 0$ (Remark 1.4)
   • Explore results under other assumptions
4. Generalizations:
   • Extend to three-prime Goldbach problem (such as similar results in Matomäki-Shao 14)
   • Study Goldbach prime gaps in arithmetic progressions
5. Technical Development:
   • Improve distribution level theorem (such as using Matomäki's exponential sum estimates, Remark 2.2)
   • Develop more general framework independent of W-trick
   • Study precise asymptotics of variational problem $M_k(D)$

In-Depth Evaluation

Strengths

1. Methodological Innovation:
   • Symmetry Observation: Simple but crucial observation (reversing roles of $f_1, f_2$) brings substantial improvement, reflecting deep insight
   • Avoiding W-trick: Develop evaluation techniques independent of Maynard and Tao, better suited to Goldbach constraints
   • Strongly Admissible Tuples: New concept with explicit construction (using squares and coprimality to 30) shows originality
2. Theoretical Depth:
   • Systematic Comparison: First in-depth comparison of two major sieve methods on the same problem, revealing essential differences
   • Barrier Identification: Proposition 5.4 precisely characterizes the variational problem barrier of Maynard-Tao method
   • Technical Completeness: From distribution level theorem (Section 2) to singular series averages (Section 3) to applications of both methods, forms complete technical chain
3. Result Significance:
   • Approximately 6.7% improvement over Tsuda, substantial progress in this type of problem
   • First establishment of absolutely bounded gaps for Goldbach primes (though constant is large)
   • All results hold for "almost all" even integers, with exception set controlled precisely ($O(X(\ln X)^{-A})$)
4. Writing Quality:
   • Clear structure: introduction, preliminaries, main lemmas, applications of two methods well-organized
   • Sufficient motivation: each technical choice clearly explained
   • Comprehensive literature review: relationships with related work thoroughly discussed

Weaknesses

1. Insufficient Numerical Optimization:
   • Authors explicitly note $H = 56250000$ is not optimized
   • No quadratic programming used to optimize sieve weights
   • No computer-aided search, actual $H(k)$ should be significantly reducible
2. Variational Problem Solution:
   • Proposition 5.4 provides upper bound but no meaningful lower bound for $M_k(D)$
   • Maynard-type argument (solving Euler-Lagrange equations) only gives heuristic choice
   • Lacks rigorous proof of $M_k$ monotonicity (though claimed $M_1 \leq M_2 \leq \cdots$)
3. Method Limitation Analysis:
   • Equation (53) proposes more general setting but not developed
   • Lacks concrete suggestions for breaking through variational barrier
   • Does not explore parity barrier effects for $k < 5$ (only cites Tao's blog)
4. Technical Details:
   • Remark 5.2 mentions similar results for $\Lambda_{Q,\text{HB}}$ are "too messy to present", lacks complete proof
   • Some lemmas (such as Lemma 2.8) directly cited from literature without proof sketches
   • Construction of strongly admissible tuples (Lemma 5.8) is clever but lacks systematic approach

Impact

1. Theoretical Contribution:
   • Advances refined theory of Goldbach prime distribution
   • Provides exemplar for sieve method comparison research
   • Strongly admissible tuple concept may have independent value
2. Methodological Impact:
   • Technique for avoiding W-trick applicable to other constrained prime problems
   • Distribution level symmetry exploitation may inspire other improvements
   • Variational problem barrier identification guides future research
3. Limitations:
   • Results still far from conjectured value ($\Xi = 0$)
   • Bounded gap constant too large for practical application
   • Method appears difficult to improve further (unless major distribution level improvement)
4. Reproducibility:
   • Theoretical proofs complete and verifiable
   • Numerical computation (Lemma 5.9) uses Mathematica, reproducible
   • No code implementation, but acceptable for pure mathematics paper

Applicable Scenarios

1. Direct Applications:
   • Study distribution density of Goldbach primes
   • Analyze properties of prime sets under additive constraints
   • Improve quantitative versions of Goldbach conjecture
2. Method Generalization:
   • Other additive prime problems (such as three-prime Goldbach, Waring-Goldbach)
   • Constrained prime distribution in arithmetic progressions
   • Polynomial prime value problems
3. Technical Reference:
   • Problems requiring comparison of different sieve methods
   • Avoiding W-trick when discriminant has large prime factors
   • Problems involving singular series averages
4. Theoretical Research:
   • Study of sieve method capabilities and limitations
   • Applications of variational problems in number theory
   • Refinements of distribution level theorems

References

This paper cites the following key references:

1. Bombieri-Davenport (1966) 3: Pioneering work on small prime gaps
2. Goldston-Pintz-Yıldırım (2009) 5: GPY method
3. Maynard (2015) 15: Breakthrough on small prime gaps
4. Polymath (2014) 22: Variant of Selberg sieve and bounded prime gaps
5. Tsuda (2024) 25: Prior work on small Goldbach prime gaps
6. Goldston-Yıldırım (2007) 6: Simplification of Bombieri-Davenport method
7. Matomäki-Shao (2017) 14: Almost twin prime version of Vinogradov's theorem
8. Gallagher (1976) 4: Prime distribution in short intervals
9. Montgomery-Vaughan (1975) 17: Exceptional sets in Goldbach problem

Overall Assessment: This is a high-quality analytic number theory paper achieving substantial progress on the Goldbach prime gap problem, with methodological innovations and deep theoretical analysis. Main contributions include improving known bounds, revealing method limitations, and establishing bounded gap existence. Although distance remains from ultimate goal ($\Xi = 0$ or existence of $\Xi^*$), it clearly indicates directions and barriers for future research. Suitable for number theory researchers, particularly experts in sieve methods and prime distribution.