Assuming the Riemann hypothesis, we establish lower bounds for large oscillations of the iterated argument functions Sn(t) of the Riemann zeta function. As a corollary, we prove that, for every fixed integer n≥0, Sn(t)=Ω±((log2t)n+1/2(logt)1/2(log3t)1/2). For even n, this recovers two results of Chirre and Mahatab in 2021. For odd n, our result strengthens theirs by establishing the missing direction of oscillation with the same order of magnitude.
Introduction
Throughout this paper, we denote the Riemann zeta function by ζ(s), where s=σ+it is a complex variable. For 1/2≤σ≤1 and t>0, define
S(σ,t):=π1argζ(σ+it),
where the argument is obtained by continuous variation along straight line segments joining the points 2, 2+it and σ+it, assuming that the segment from σ+it to 2+it has no zeros of ζ(s), and with the convention that argζ(2)=0. If this path has zeros of ζ(s) (including the endpoint σ+it), define
S(σ,t)=2S(σ,t+0)+S(σ,t−0).
Set S0(σ,t)=S(σ,t). Then, for integer n≥1, define
Sn(σ,t)=∫0tSn−1(σ,τ)dτ+δn,σ,
where the constant δn,σ is given by Fujii [11]. More precisely, for n=2k with k≥1,
The function S(t) is closely related to the distribution of the non-trivial zeros of the Riemann zeta function. The classical Riemann–von Mangoldt formula states that
N(t)=2πtlog(2πet)+87+S(t)+O(t−1).
where N(t) denotes the number of zeros ρ=β+iγ of ζ(s) with 0<γ≤t, counted with multiplicity, where zeros with ordinate t are counted with half weight. In particular, for 0<h≤1, we have
N(t+h)−N(t)=2πhlogt+S(t+h)−S(t)+O(h+t−1).
This relation shows that the oscillations of S(t) are intimately connected with the local distribution of zeros of ζ(s). More generally, the iterated functions Sn(t) describe successive integrals of these fluctuations. Understanding their large values and oscillations is therefore a natural problem in the study of the Riemann zeta function.
The study of large values of S(t) has a long history. In 1977, Montgomery [15] established, assuming the Riemann hypothesis (RH), the following Ω±-result:
S(t)=Ω±((log2t)1/2(logt)1/2).
Here and throughout this paper, we write logj for the j-th iterated logarithm. A heuristic prediction for the true order of magnitude of S(t) was proposed by Farmer, Gonek and Hughes [10]. Based on random matrix theory, they conjectured that
t→∞limsup(logt)1/2(log2t)1/2S(t)=π21.
This conjecture suggests that large values of S(t) may be substantially larger than the currently known lower bounds.
Tsang [16] subsequently obtained several refinements of the classical Ω-results. More precisely, he proved that
In addition, Tsang obtained unconditional estimates for S1(t), with different orders in the two directions. In a subsequent paper, Tsang [17] further improved Ω+-result for S1(t).
A significant development came from the resonance method. In 2018, Bondarenko and Seip [2 Theorem 2] proved, under RH, that
These results introduced an additional factor of (log3t)1/2 compared with the earlier conditional work, although the bound for S(t) did not distinguish between its positive and negative values.
Using a modified version of the resonance method in [2], Chirre [7] investigated large values of Sn(σ,t) near the critical line. Assuming RH, when σ∈[1/2,1/2+1/(log2T)], he obtained lower bounds of order, for sufficiently large T,
(log2T)σ+n(logT)1−σ(log3T)σ.
When n≡1(mod4), he established an Ω+-result for Sn(σ,t), whereas in the other cases, a corresponding Ω-result is available. In 2021, Chirre and Mahatab subsequently made two important contributions. Specifically, in [8] they obtained that
S(t)=Ω±((log2t)1/2(logt)1/2(log3t)1/2).
This result complements the missing negative direction in [2]. On the other hand, Chirre and Mahatab [9] extended this approach to the iterated functions Sn(σ,t) near the critical line. Assuming RH, they established large oscillations of Sn(σ,t+h)−Sn(σ,t) for every fixed integer n≥1. Their result gives both directions when n is odd, but only one prescribed direction when n is even. Consequently, their results for Sn(t) are two-sided for even n, whereas for odd n only one direction is obtained at this order of magnitude. They also derived certain unconditional one-sided consequences for higher iterates.
More recently, Inoue, Kobayashi and Toma [13] further studied explicit extreme values of S(t+h)−S(t) in short intervals under RH. Their results also have applications to large and small gaps between consecutive zeros of the Riemann zeta function. For more developments and details about the argument of ζ(s) and its iterates, we recommend [3, 4, 5, 6, 11, 12] and the references therein.
Motivated by these developments, we investigate whether large oscillation estimates of Chirre and Mahatab [9] can be strengthened to both directions for every fixed integer n≥1. Our main result gives an affirmative answer under RH, without imposing any restriction on the parity of n.
Theorem 1.1. Assume RH. Let n≥1 be a fixed integer and let 0<β<1 be fixed. Then there exists a constant cn,β>0 such that, for all sufficiently large T and
Corollary 1.2. Assume RH. Let n≥0 be a fixed integer and let 0<β<1 be fixed. Then there exists a constant cn,β>0 such that, for all sufficiently large T, we have
We now compare our results with the work of Chirre and Mahatab. In fact, [9 Theorem 1] establishes two-sided large oscillations of Sn(t+h)−Sn(t) when n is odd, but only one prescribed direction when n is even. Theorem 1.1 removes this restriction and establishes both directions for every fixed integer n≥1, with the same order of magnitude. On the other hand, for Sn(t), Corollary 1.2 recovers [9 Corollary 2] when n is even and σ=1/2. When n is odd, it strengthens their result by establishing the missing direction. Furthermore, when n=0, Corollary 1.2 recovers [8 Theorem 1].
Although our results are stated on the critical line for simplicity, the same method yields the corresponding estimates uniformly for 1/2≤σ≤1/2+(log2T)−1, with Sn(t) replaced by Sn(σ,t).
The choice of the kernel K(u), defined in (1), is inspired by Jaskari [14]. The resulting convolution formula, established in Proposition 2.3, plays a crucial role in the proof of Theorem 1.1. In contrast to [9 Proposition 8], our approach avoids determining which of two finite sums yields the main term based on the parity of n. This is achieved by suitably choosing the phase of the sine function in K(u). Since in(−i)n−1=i, this choice guarantees that, after taking the imaginary part, the second sum always appears with coefficient ±1/2, independent of n.
Preliminary tools
A convolution formula
For h∈R and a function F, we write
ΔhF(t)=F(t+h)−F(t−h).
The following two lemmas, due to Chirre and Mahatab [9], play a crucial role in establishing the convolution formula we require.
Lemma 2.1 (Chirre-Mahatab [9], Lemma 6). Assume RH. Let 0<β<1 be a fixed number. Let α>0, H∈R and T be sufficiently large. Then for 1/2<ν≤2, 0≤h≤1
Note that K(u) is always positive. More precisely, 1≤K(u)≤3.
Proposition 2.3.Assume RH. Let n≥1 be a fixed integer and let 0<β<1 and 0<γ≤1/2 be fixed. Let T be sufficiently large. Then for 0≤h≤1 and Tβ≤t≤TlogT, we have
Here, we use the linear combination (1). Furthermore, when 0<γ≤1/2, we have ωm(γlog2T,−log2T)=0 for all m≥2. Multiplying both sides of (2) by (v−1/2)n−1 and integrating over v∈[1/2,2] gives
Here we also use the following trivial upper bound that, for j=1,2,
∣aj(m)∣≪hlogmlog2T.(4)
Multiplying by the factor in/(π(n−1)!) and taking the imaginary part of both sides, we complete the proof of Proposition 2.3. □
The resonator and some estimates
We now construct the required resonator. Indeed, the resonator we use here coincides with the one chosen in [8], both following the idea of [1, 2, 7]. For completeness, we recall its construction and the estimates needed in our proof.
Fix β∈(0,1) and set κ=(1−β)/2, so that κ+β<1. For sufficiently large T, set
N=⌊Tκ⌋.
Let P denote the set of primes p satisfying
elogNlog2N<p≤exp((log2N)1/8)logNlog2N.
Next, we define the multiplicative function f(n) to be supported on the set of square-free numbers, with values for p∈P as
f(p)=log3NlogNlog2Np(logp−log2N−log3N)1.
Furthermore, if p∈/P, set f(p)=0.
For k∈{1,…,⌊(log2N)1/8⌋}, we define the set Pk of prime numbers as follows
Pk:={p:eklogNlog2N<p≤ek+1logNlog2N}
and
Mk:={n∈supp(f):n has at least k2log3N3logN prime divisors in Pk}.
Furthermore, define
M:=supp(f)∖k=1⋃⌊(log2N)1/8⌋Mk.
Let L be the set of integers ℓ such that
[(1+T1)ℓ,(1+T1)ℓ+1]∩M=∅.
Then, we let mℓ be the minimum of [(1+T−1)ℓ,(1+T−1)ℓ+1]∩M for ℓ∈L. Define the set
M′:={mℓ:ℓ∈L}.
and define
r(mℓ):=(1−T−1)ℓ−1≤n≤(1+T−1)ℓ+2n∈M∑f(n)21/2
for every mℓ∈M′. Moreover, the resonator is defined by
R(t):=m∈M′∑r(m)m−it.
We conclude this section with some estimates. The Cauchy-Schwarz inequality implies that
∣R(t)∣2≤R(0)2≪Tκk∈M∑f(k)2.(5)
Furthermore, set Φ(y):=e−y2/2 as in [1]. It follows from [2 Lemma 5] (also see [8 Proposition 4]) that
∫−∞∞∣R(t)∣2Φ(Tt)dt≪Tk∈M∑f(k)2.(6)
We will make frequent use of (5) and (6) in the subsequent argument. Finally, owing to the work of Bondarenko and Seip [2], we state the following lemma:
In this section, we combine the resonance method with Proposition 2.3 to prove Theorem 1.1. The case h=0 is trivial, so we assume that h∈(0,(2log2T)−1) throughout the proof. Set γ=1/8. First, by Proposition 2.3, we have
Here, in the last step, we make use of (6) and the fact that both ∣R(t)∣2 and Φ(t/T) are non-negative.
We first consider the second term on the right-hand side of (7), namely the one associated with a2(m), as this contributes the main term. To this end, we wish to expand the integral over R, so that Lemma 2.4 can be applied. By the definition of ωm, a2(m)=0 unless (logT)3/4≤m≤(logT)5/4. Thus, combining this with (4) and (5), we obtain
On the other hand, we seek an upper bound for the first term on the right-hand side of (7). By the definition of ωm, we have a1(m)=0 whenever m>(logT)1/4. Hence, the sum runs over 2≤m≤(logT)1/4. Together with (4) and (6), this gives
Finally, we adjust the parameters to obtain the desired result. Recall that
Δh/2Sn(v)=Sn(v+h/2)−Sn(v−h/2).
Putting t=v−h/2, we have
Δh/2Sn(v)=Sn(t+h)−Sn(t).
Now, fix 0<β<β1<1 and put T1=T/(4logT). We apply the preceding estimate with T replaced by T1, β replaced by β1, and h replaced by h/2. Since 0<h≤(log2T)−1, we have
where we used logjT1∼logjT for j=1,2,3. Since the estimate holds for each ε∈{−1,1}, the proof of Theorem 1.1 is complete.
Proof of Corollary 1.2
Set β′=(1+β)/2. Applying Theorem 1.1 with n replaced by n+1, T replaced by T/2, we choose h=(log2(T/2))−1, which lies in the admissible range of the theorem. Then, we obtain
Since β′>β, we have (T/2)β′≥Tβ and T/2+h≤T for all sufficiently large T. Hence, [t,t+h]⊆[Tβ,T], which completes the proof of Corollary 1.2.
AI Declaration
The authors used ChatGPT solely for language revision in this work and assume full responsibility for the mathematical arguments, results, and final text.
References
[1]A. Bondarenko and K. Seip, Large greatest common divisor sums and extreme values of the Riemann zeta function, Duke Math. J., 166 (2017), pp. 1685–1701.
[2]A. Bondarenko and K. Seip, Extreme values of the Riemann zeta function and its argument, Math. Ann., 372 (2018), pp. 999–1015.DOI
[3]H. M. Bui, S. J. Lester, and M. B. Milinovich, On Balazard, Saias, and Yor’s equivalence to the Riemann Hypothesis, J. Math. Anal. Appl., 409 (2014), pp. 244–253.DOI
[4]E. Carneiro, V. Chandee, and M. B. Milinovich, Bounding S(t) and S₁(t) on the Riemann hypothesis, Math. Ann., 356 (2013), pp. 939–968.
[5]E. Carneiro and A. Chirre, Bounding Sₙ(t) on the Riemann hypothesis, Math. Proc. Camb. Philos. Soc., 164 (2018), pp. 259–283.DOI
[6]E. Carneiro, A. Chirre, and M. B. Milinovich, Bandlimited approximations and estimates for the Riemann zeta-function, Publ. Mat., 63 (2019), pp. 601–661.DOI
[7]A. Chirre, Extreme values for Sₙ(σ,t) near the critical line, J. Number Theory, 200 (2019), pp. 329–352.
[8]A. Chirre and K. Mahatab, Large oscillations of the argument of the Riemann zeta-function, Bull. Lond. Math. Soc., 53 (2021), pp. 1776–1785.arxiv.org/abs/1904.11051
[9]A. Chirre and K. Mahatab, Large values of the argument of the Riemann zeta-function and its iterates, J. Number Theory, 225 (2021), pp. 240–259.arxiv.org/abs/2006.04288
[10]D. W. Farmer, S. M. Gonek, and C. P. Hughes, The maximum size of L-functions, J. reine angew. Math., 609 (2007), pp. 215–236.DOI
[11]A. Fujii, On the zeros of the Riemann zeta function, Comment. Math. Univ. St. Pauli, 51 (2002), pp. 1–17.DOI
[12]S. Inoue, Extreme values for iterated integrals of the logarithm of the Riemann zeta-function, Acta Arith., 205 (2022), pp. 97–119.arxiv.org/abs/2009.04099
[13]S. Inoue, H. Kobayashi, and Y. Toma, Explicit extreme values of the argument of the Riemann zeta-function, Math. Proc. Camb. Philos. Soc., 181 (2026), pp. 1149–1163.
[14]M. Jaskari, Applying the resonance method to Re(e^{-iθ} log ζ(σ + it)), Funct. Approx. Comment. Math., 70 (2024), pp. 263–275.DOI
[15]H. L. Montgomery, Extreme values of the Riemann zeta function, Comment. Math. Helv., 52 (1977), pp. 511–518.DOI
[16]K.-M. Tsang, Some Ω-theorems for the Riemann zeta-function, Acta Arith., 46 (1986), pp. 369–395.DOI
[17]K.-M. Tsang, An improved Ω₊ theorem for S₁(t), Q. J. Math., 42 (1991), pp. 501–513.DOI