Papers
Topics
Authors
Recent
2000 character limit reached

Baker-Montgomery Conjecture on L-Function Zeros

Updated 9 November 2025
  • Baker-Montgomery Conjecture is a hypothesis on the asymptotic behavior of real zeros of L'(s,χ₋d), predicting their increase at a rate proportional to log log |d|.
  • The conjecture connects analytic number theory with random matrix theory by employing Selberg's short Dirichlet polynomial approximations and probabilistic models to estimate zero distributions.
  • Recent work by Lamzouri and Nath has nearly achieved an unconditional resolution up to a polylogarithmic factor, highlighting both upper and lower bounds for real zeros in the critical interval.

The Baker-Montgomery conjecture concerns the asymptotic behavior of the number of real zeros of the derivative of quadratic Dirichlet LL-functions, specifically L(s,χd)L'(s, \chi_d), within the critical interval [1/2,1][1/2,1]. Given its focus on fundamental discriminants and the distribution of zeros near the critical line, the conjecture is a central problem in the analytic study of LL-functions and their derivatives, linking zero distribution phenomena to random matrix theory and sieving methods. Recent advances, notably by Lamzouri and Nath, have nearly resolved the conjecture unconditionally up to a polylogarithmic factor, with a full resolution conditional on well-motivated analytic hypotheses.

1. Formal Statement of the Baker–Montgomery Conjecture

Let dd be a fundamental discriminant and χd\chi_d the associated primitive quadratic Dirichlet character. Define Rd(σ1,σ2)R_d(\sigma_1,\sigma_2) as the number of real zeros of L(s,χd)L'(s,\chi_d) in the interval [σ1,σ2][\sigma_1,\sigma_2]. The conjecture posits:

For almost all fundamental discriminants dd \to \infty,

>Rd(1/2,1)Cloglogd>> R_d(1/2,1) \sim C \log\log|d| >

for some positive constant CC; in particular, Rd(1/2,1)loglogdR_d(1/2,1) \asymp \log\log|d|.

This predicts that the number of real zeros of L(s,χd)L'(s,\chi_d) in [1/2,1][1/2,1] for almost all dd grows like the double logarithm of d|d|.

2. Definitions and Foundational Objects

  • Quadratic Dirichlet Character χd(n)\chi_d(n): The primitive character modulo d|d| defined via the Kronecker symbol (d/n)(d/n).
  • Fundamental Discriminant dd: The discriminant of the quadratic field Q(d)\mathbb{Q}(\sqrt{d}), with d1(mod4)d \equiv 1 \pmod{4} and squarefree, or d=4md=4m, m2,3(mod4)m\equiv2,3\pmod{4} and squarefree.
  • Dirichlet LL-function:

L(s,χd)=n=1χd(n)ns,s>1L(s,\chi_d) = \sum_{n=1}^\infty \frac{\chi_d(n)}{n^{s}},\qquad \Re s>1

Analytically continued to C\mathbb{C}; its derivative is L(s,χd)=ddsL(s,χd)L'(s, \chi_d) = \frac{d}{ds}L(s,\chi_d).

3. Unconditional Results: Upper and Lower Bounds

Upper Bound (Lamzouri–Nath):

Let ν(x)\nu(x)\to\infty as xx\to\infty, and D(x)={D(x) = \{fundamental discriminants dd in (x/2,x](x/2, x] in a fixed progression}\}, D(x)x|D(x)| \asymp x. Theorem 1.1 asserts:

Rd(12+ν(x)logx,1)(loglogx)(logloglogx)R_d\left(\frac{1}{2}+\frac{\nu(x)}{\log x}, 1\right) \ll (\log\log x) (\log\log\log x)

for almost all dD(x)d \in D(x). Thus, outside a negligible exceptional set, L(s,χd)L'(s,\chi_d) has at most O((loglogd)(logloglogd))O((\log\log |d|)(\log\log\log |d|)) real zeros in [12+ν(d)logd,1][\frac{1}{2}+\frac{\nu(|d|)}{\log|d|}, 1].

Proof strategy summary:

  • Approximate LL(s,χd)-\frac{L'}{L}(s,\chi_d) by short Dirichlet polynomials (Selberg approximation).
  • Develop a probabilistic random model by replacing χd(n)\chi_d(n) with i.i.d. random variables.
  • Compare moments via discrepancy bounds.
  • Use a discretized net {zj=1/2+3j}\{z_j = 1/2+3^{-j}\}; control Ld(zj)L_d(z_j) uniformly.
  • Apply Jensen’s formula over concentric circles to bound the count of real zeros.

Lower Bound (Klurman–Lamzouri–Munsch):

For almost all dD(x)d\in D(x),

Rd(1/2,1)loglogdlog4dR_d(1/2,1) \gg \frac{\log\log|d|}{\log_4|d|}

Moreover, all such zeros lie in [1/2+1/(logx)1/5,1][1/2+1/(\log x)^{1/5},1].

Combined Order of Magnitude (Corollary 1.2):

For almost all dd,

Rd(12+ν(x)logx,1)=(loglogx)(logloglogx)θR_d\left(\frac{1}{2}+\frac{\nu(x)}{\log x},1\right) = (\log\log x)(\log\log\log x)^\theta

for some θ1|\theta|\leq 1, exhibiting the optimal shape up to a polylogarithmic factor.

4. Conditional Results and Hypotheses

Assuming a mild zero-repulsion hypothesis on the low-lying zeros of quadratic Dirichlet LL-functions—precisely, that for each dD(x)d\in D(x) there exists ν(x)\nu(x)\rightarrow\infty so that L(s,χd)L(s,\chi_d) has no zeros in a disk centered at 1/2+ν(x)/logx1/2+\nu(x)/\log x of suitable small radius (Hypothesis LdL_d)—a full resolution of the conjecture is obtained up to the logloglogd\log\log\log|d| factor:

Theorem 1.3 (Conditional): Assuming Hypothesis LdL_d for almost all dD(x)d \in D(x),

  • Rd(1/2,1)(loglogx)(logloglogx)R_d(1/2,1) \ll (\log\log x)(\log\log\log x)
  • Rd(1/2,1/2+ν(x)/logx)=o(Rd(1/2+ν(x)/logx,1))R_d(1/2,\,1/2+\nu(x)/\log x) = o(R_d(1/2+\nu(x)/\log x, 1))

When GRH and the one-level density conjecture of Katz and Sarnak hold (established for support up to (2,2)(-2,2) by Özlük and Snyder), Hypothesis LdL_d is satisfied for almost all dd.

Key technical tools employed include:

  • Orthogonality relations for χd(n)\chi_d(n) (Lemma 2.1)
  • Zero-density theorems in the critical strip (Lemma 2.7)
  • Selberg's short Dirichlet polynomial bounds (Lemma 3.1)
  • Discrepancy bounds versus random models (Proposition 4.2)
  • Moment estimates for L(s)L(s) near and away from $1/2$ (Lemmas 4.3, 5.1, Proposition 5.2)

5. Error Terms, Polylogarithmic Factors, and Localization

Error Source and Estimates

  • Exceptional set for Theorem 1.1: size O(xexp(cν(x))+x(loglogx)3)O\left(x\cdot\exp(-c\nu(x)) + x \cdot (\log\log x)^{-3}\right), hence for slowly diverging ν(x)\nu(x), nearly all dd are included.
  • For Theorem 1.3: the exceptional proportion is O(logν(x)/ν(x))O(\sqrt{\log\nu(x)/\nu(x)}).
  • The logloglogx\log\log\log x factor originates from:
    • Multiple applications of Jensen’s formula on nets of points (with JloglogxJ\sim\log\log x), each incurring an additive logloglogx\log\log\log x term;
    • Discrepancy error terms near the boundary of the critical interval, scaling like (VjloglogVj/logd)1/2(V_j \log\log V_j/\log d)^{1/2}, with Vj=3jV_j = 3^j.

Support for "All Zeros to the Right":

Under Hypothesis LdL_d, almost all real zeros of L(s,χd)L'(s, \chi_d) in [1/2,1][1/2,1] satisfy

Rd(1/2,1/2+ν/logx)=o(Rd(1/2+ν/logx,1)),R_d(1/2,1/2+\nu/\log x) = o(R_d(1/2+\nu/\log x,1)),

hence asymptotically all such zeros are confined to [1/2+ν/logx,1][1/2+\nu/\log x,1]. Methods employ moment bounds on L(s)L(s) near s0=1/2+ν/logxs_0=1/2+\nu/\log x, Jensen's formula on nested circles, hole-size control by the zero-free disk, and Cauchy-integral mean-value estimates.

6. Remaining Obstacles and Steps Toward a Full Unconditional Proof

Unconditional results leave open the “near-critical” interval [1/2,1/2+ν/logd][1/2,1/2+\nu/\log |d|], offering no control over zeros in this sliver, and have a O(logloglogd)O(\log\log\log|d|) slack from the predicted asymptotic. Removal of the last polylogarithmic factor and full control as σ1/2\sigma\searrow 1/2 would require:

  • Stronger uniform discrepancy estimates for LL(s,χd)-\frac{L'}{L}(s,\chi_d) for 1/2<σ<1/2+O(1/logx)1/2 < \sigma < 1/2 + O(1/\log x),
  • Sharper understanding of large-value behavior of LL-\frac{L'}{L} very close to the $1/2$-line,
  • Improved zero-density estimates within the near-critical strip.

As of Lamzouri and Nath’s work, methods reach within a polylogarithmic factor of confirming the conjecture, with a full conditional resolution when natural hypotheses on the distribution of low-lying zeros are assumed.

The Baker–Montgomery conjecture is part of a broader program investigating the vertical and horizontal distribution of zeros of LL-functions and their derivatives. The moment-discrepancy methodology, as refined by Lamzouri and Nath, uses a random model to approximate LL-function behavior and integrates zero-density, orthogonality, and polynomial approximation techniques. The work interacts with the one-level density conjecture and Katz–Sarnak random matrix models, providing strong evidence for deep connections between analytic multiplicities, random matrix statistics, and arithmetic geometry.

A plausible implication is that refinements of random-model comparisons and explicit moment computations for Dirichlet LL-functions may generalize to higher-degree LL-functions or to derivatives of automorphic LL-functions, strengthening the bridge between analytic number theory and statistical mechanics models.


Selected works:

  • R. C. Baker & H. L. Montgomery, “Oscillations of quadratic L-functions,” 1989.
  • O. Klurman, Y. Lamzouri & M. Munsch, “Sign changes of short character sums and real zeros of Fekete polynomials,” (Klurman et al., 4 Mar 2024).
  • J. B. Conrey & K. Soundararajan, “Real zeros of quadratic Dirichlet L-functions,” Invent. Math. 150 (2002).
  • A. E. Özlük & C. Snyder, “On the distribution of the nontrivial zeros of quadratic L-functions close to the real axis,” Acta Arith. 91 (1999).
Definition Search Book Streamline Icon: https://streamlinehq.com
References (1)

Whiteboard

Topic to Video (Beta)

Follow Topic

Get notified by email when new papers are published related to Baker-Montgomery Conjecture.