June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

International Journal of Number Theory

Vol. 19, No. 6 (2023) 1205–1228
c World Scientific Publishing Company
by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

DOI: 10.1142/S1793042123500598

On the error term in the explicit formula

of Riemann–von Mangoldt

Michaela Cully-Hugill∗ and Daniel R. Johnston†

School of Science, UNSW Canberra, Australia
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

∗[email protected]
†[email protected]

Received 23 November 2021

Revised 28 October 2022
Accepted 23 March 2023
Published 16 May 2023

We provide an explicit O(x log x/T ) error term for the Riemann–von Mangoldt formula
by making results of Wolke [On the explicit formula of Riemann–von Mangoldt, II,
J. London Math. Soc. 2(3) (1983) 406–416] and Ramaré [Modified truncated Perron
formulae, Ann. Math. Blaise Pascal 23(1) (2016) 109–128] explicit. We also include
applications to primes between consecutive powers, the error term in the prime number
theorem and an inequality of Ramanujan.

Keywords: Distribution of primes; Riemann von Mangoldt formula; gaps between primes;
Riemann zeta-function.

Mathematics Subject Classification 2020: 11N05, 11M26, 11Y70

1. Introduction
1.1. Background
Let
 
ψ(x) = Λ(n) = log p
n≤x pk ≤x

be the Chebyshev prime-counting function over prime powers pk , with integers

k ≥ 1. The (truncated) Riemann–von Mangoldt formula is written as
 xρ
ψ(x) = x − + E(x, T ), (1.1)
ρ
ρ=β+iγ
|γ|≤T

where the sum is over all non-trivial zeros ρ = β + iγ of the Riemann-zeta function
ζ(s) that have |γ| ≤ T , and E(x, T ) is an error term.

∗ Corresponding author.

1205
 June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

1206 M. Cully-Hugill & D. R. Johnston

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

Several authors have provided estimates for E(x, T ). With reasonable conditions
on x and T , the standard bound given in the literature (e.g. [11, Chap. 17]) is
 
x log2 x
E(x, T ) = O , (1.2)
T
which in 1983 was improved by Goldston [17] to
 
x log x log log x
E(x, T ) = O . (1.3)
T
With more careful bounding, one can go further. Several authors [24, Corol-
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

lary 5.3; 37, Theorem 2.3, p. 219; 34, Theorem 1.1] prove results which imply a
bound of
 
x log x
E(x, T ) = O . (1.4)
T
Wolke [38] and Ramaré [34] also claimed to prove averaged versions of the Riemann–
von Mangoldt formula with O(x/T ) error terms. However, through correspondence
with Ramaré, several errors in these works have been uncovered.
In 2016, Dudek [13, Theorem 1.3] gave an explicit version of (1.2). Namely, he
proved the following theorem.

Theorem 1.1 (Dudek). Take 50 < T < x for half an odd integer x > e60 .
Then
 
2x log2 x
E(x, T ) = O∗ .
T
The first author [7, Theorem 2] recently improved on Dudek’s result by mak-
ing (1.3) explicit. In this paper, we give an explicit O(x log x/T ) bound for E(x, T ),
thereby making (1.4) explicit. This is done by reworking the papers of Wolke [38]
and Ramaré [34]: we optimize parts of their proofs and avoid the errors related to
their averaging arguments. This approach gives significantly better explicit bounds
for E(x, T ) than in the previous works.
Our overarching approach is to split E(x, T ) into two smaller error terms, say
E1 (x, T ) and E2 (x, T ). We rework [34, Theorem 1.1] to bound E1 (x, T ) in a general
way that can be applied to other arithmetic functions besides ψ(x). Then, we rework
the proof of [38, Theorem 2] and use explicit zero-free regions for ζ(s) to reach an
explicit estimate of the form E2 (x, T ) = O(x/T ). Such an estimate for E2 (x, T )
becomes insignificant compared to that for E1 (x, T ) for large x.
To demonstrate the usefulness of our results, we use our bounds for E(x, T ) to
improve the main theorems in [7, 32].

1.2. Statement of main results

Our main result is as follows.
June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

The error term in the explicit formula of Riemann–von Mangoldt 1207

Theorem 1.2. For any α ∈ (0, 1/2] there exist constants M and xM such that for
by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

max{51, log x} < T < (xα − 2)/2,

 xρ  
∗ x log x
ψ(x) = x − +O M (1.5)
ρ T
|γ|≤T

for all x ≥ xM . Some admissible values of xM , α and M are (40, 1/2, 5.03) and
(103 , 1/100, 0.5597), with more given in Table A.2 in the Appendix.
Using Theorem 1.2 we are able to obtain the following result.

Theorem 1.3. There is at least one prime between n140 and (n + 1)140 for all
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

n ≥ 1.
This improves upon [7, Theorem 1] by the first author, which asserts that there
is always a prime between consecutive 155th powers.
Theorem 1.2, combined with other recent results, also allows us to improve the
error term in the prime number theorem for large x (cf. [32, Theorem 1]).

Theorem 1.4. Let R = 5.5666305. For each {X, A, B, C, 0 } in Table 1 we have

   B   
 ψ(x) − x  log x log x
 ≤A exp −C ,
 x  R R
and for all log x ≥ X,
|ψ(x) − x| ≤ 0 x.
The values of 0 in Table 1 are 40–80% smaller than those in [32, Table 1].
Replacing A with A1 = A + 0.1 gives a corresponding expression for θ(x) (see
Corollary 7.2). It should be noted that the methods used to prove Theorem 1.4
have been expanded on in recent preprints [15, 22].

Table 1. Values of X, A, B, C and 0 for Theorem 1.4. Here, σ is

a parameter that appears in the proof of Theorem 1.4. The entry
for X = 3600 is specifically included for Corollary 1.5.
X σ A B C 0
1000 0.980 269.1 1.520 1.893 6.89 · 10−6
2000 0.984 264.8 1.516 1.914 3.48 · 10−10
3000 0.986 264.3 1.514 1.925 1.42 · 10−13
3600 0.988 275.2 1.512 1.936 2.04 · 10−15
4000 0.988 266.5 1.5212 1.936 1.61 · 10−16
5000 0.990 350.4 1.510 1.946 4.74 · 10−19
6000 0.990 267.8 1.510 1.946 1.83 · 10−21
7000 0.990 266.9 1.510 1.946 1.38 · 10−23
8000 0.990 266.9 1.510 1.946 1.44 · 10−25
9000 0.992 280.5 1.508 1.957 1.30 · 10−27
10000 0.992 268.6 1.508 1.957 2.06 · 10−29
 June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

1208 M. Cully-Hugill & D. R. Johnston

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

By repeating the argument on [32, pp. 877–880], we obtain the following appli-
cation to an inequality of Ramanujan on the prime counting function π(x).

Corollary 1.5. For all x ≥ exp(3604) we have

ex x
π(x)2 < π .
log x e
Corollary 1.5 improves [32, Theorem 2] by a factor of exp(311).

1.3. Outline of paper

Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

We begin in Sec. 2 by proving an explicit and simplified version of a truncated

Perron formula due to Ramaré [34]. In Sec. 3, we state a number of zero-free regions
from the literature that will be required throughout the paper. In Sec. 4, we make
explicit an argument of Wolke [38]. These results are combined in Sec. 5 to prove
Theorem 1.2. Applications of these results (Theorems 1.3 and 1.4) are in Secs. 6
and 7.

2. An Explicit Truncated Perron Formula

In this section, we prove the following error estimate for the truncated Perron
formula. This result is a self-contained variant of [34, Theorem 1.1], which we have
simplified and optimized for our purposes.

Theorem 2.1. Let F (s) = n≥1 an /ns be a Dirichlet series over complex s, and
κ > 0 be a real parameter√chosen larger than κa the abscissa of absolute convergence
of F (s). Also let θ = 2/( π 2 + 4+π). For any T > 0, x ≥ 1, κ > κa , and λ ≥ θ /T,
 1 κ+iT
xs
an = F (s) ds
2πi κ−iT s
n≤x
⎛ ⎞
xκ  |an | 1 λ  eκu ⎠
+ O∗ ⎝ + |an | 2 du .
πλT nκ πT θ  /T |log(x/n)|≤u u
n≥1

Theorem 2.1 is proven using Lemma 2.2, which is a specific case of [34,
Lemma 2.2]. For both proofs, we will use the step function

1 y ≥ 0,
v(y) =
0 y < 0.

Lemma 2.2. For κ > 0 and y ∈ R we have

      
 κ +i ys  eyκ  eyκ
  |y|eyκ
 1 e   
v(y) − ds ≤ min , v(y) − arctan(1/κ ) + .
 2πi κ −i s  π|y|  π  π
(2.1)
June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

The error term in the explicit formula of Riemann–von Mangoldt 1209

Proof. For y ∈ R and K > κ we can evaluate the contour integral

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

  
κ +i K+i K−i κ −i
eys
+ + + ds = 0.
κ −i κ +i K+i K−i s
Consider the case y < 0: as K approaches infinity the third integral approaches

zero, and the two horizontal integrals are bounded by eyκ /|y|. Hence,
  
 κ +i eys  2eyκ
 
 ds ≤ .
 κ −i s  |y|
Therefore, we have
 
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

 κ +i
eys  eyκ

 1
v(y) − ds ≤ .
 2πi κ −i s  π|y|

The case y > 0 is similar, but we use a contour extended to the left, so for K < 0,
  
κ +i K+i K−i κ −i
eys
+ + + ds = 2πi.
κ −i κ +i K+i K−i s
Taking K → −∞ brings the third integral to zero, so
 
 κ +i ys  2eyκ
 e 
2πi − ds ≤ ,
 κ −i s  y
and thus, for y > 0
 
 κ +i
eys  eyκ

 1
v(y) − ds ≤ .
 2πi κ −i s  πy
The above bounds are most useful for large y. For small y we can use
κ +i κ +i 1
eys  ds  eiyt − 1
ds = eyκ + ieyκ dt.
κ −i s κ −i s −1 κ + it

The first integral is equivalent to 2i arctan(1/κ ). The second integral can be
bounded using the identity
 iyt   
 e − 1   1 iytu 
 = e du  ≤ 1.
 iyt   
0

Hence, for all y ∈ R,

   
 κ +i
eys   eyκ
  |y|eyκ
 1  
v(y) − ds ≤ v(y) − arctan(1/κ ) + .
 2πi κ −i s   π  π

Proof of Theorem 2.1. We aim to bound

 
 
 1 κ+iT
xs  −κ
 a v(T log(x/n)) − F (s) ds x
 n
2πi κ−iT s 
n≥1
 June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

1210 M. Cully-Hugill & D. R. Johnston

for any x ≥ 1, T > 0, and κ > κa > 0, where κa is the abscissa of absolute
by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

convergence of F (s). First, we take κ = κ T to reach

 
 
 1 κ+iT
xs 
 an v(T log(x/n)) − F (s) ds x−κ
 2πi κ−iT s 
n≥1
 
 |an |  1 κ+iT
x s ds  n κ
≤ v(T log(x/n)) − 
nκ  2πi κ−iT n s x
n≥1
 
 |an |  1 κ +i
dw 
 −κ T log(x/n)
= v(T log(x/n)) − eT log(x/n)w e . (2.2)
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

nκ  2πi κ −i w 
n≥1

Next, we apply Lemma 2.2. Suppose for some θ > 0 there exists a positive constant
c such that
    
1  v(y) 1  |y|
 
max min , − arctan(1/κ ) + ≤ c. (2.3)
|y|<θ π|y|  eyκ π π
Splitting the sum in (2.2) at θ and taking y = T log(x/n) in Lemma 2.2 gives
 
 
 1 κ+iT
xs  −κ
 a v(T log(x/n)) − F (s) ds x
 n
2πi κ−iT s 
n≥1
 |an | 1  |an |
≤c + , (2.4)
nκ πT nκ | log(x/n)|
T | log(x/n)|<θ T | log(x/n)|≥θ

and note that the first bound in (2.1) was used to obtain the second term in (2.4).
We then have
 |an |  |an | ∞ du
=
nκ |log(x/n)| nκ |log(x/n)| u2
T |log(x/n)|≥θ T |log(x/n)|≥θ
∞  |an | du
=
θ/T θ/T ≤|log(x/n)|≤u nκ u 2
∞  |an | du T  |an |
= − .
θ/T |log(x/n)|≤u nκ u 2 θ nκ
T |log(x/n)|<θ

Hence,
 
 
 1 κ+iT
xs  −κ
 an v(T log(x/n)) − F (s) ds x
 2πi s 
n≥1 κ−iT

∞    
1 |an | du 1 |an |
≤ + c− . (2.5)
πT θ/T | log(x/n)|≤u nκ u 2 πθ nκ
T |log(x/n)|<θ
June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

The error term in the explicit formula of Riemann–von Mangoldt 1211


Over T |log(x/n)| < θ we have |v(y)e−yκ − arctan(1/κ )/π| < 1, so we can take
by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

   √ 2
1 |y| π +4 1
c = max min ,1 + = + .
|y|<θ π|y| π 2π 2
√
This implies that if we choose θ ≤ θ := 2/( π 2 + 4 + π) then the second term
of (2.5) vanishes. Also, the integral in the first term of (2.5) will be minimized for
larger θ. Hence, we will take θ = θ . This leaves us with
 
 
 1 κ+iT
x s  κ ∞  |an | du
 a v(T log(x/n)) − F (s) ds ≤ x .
 n
2πi κ−iT 
s  πT θ /T nκ u 2

Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

n≥1 |log(x/n)|≤u

Splitting the integral from θ /T to ∞ at λ then gives the further bound

∞  |an | du  |an | ∞
du λ  x κ du
xκ κ 2
≤ xκ + |an |
θ  /T |log(x/n)|≤u n u nκ λ u2 θ  /T |log(x/n)|≤u n u2
n≥1

xκ  |an | λ  eκu
≤ + |an | du,
λ nκ θ  /T |log(x/n)|≤u u2
n≥1

from which Theorem 2.1 follows.

3. Zero-Free Regions
There are a range of explicit zero-free regions for ζ(s) in the literature. We will
combine several existing results to have an optimal zero-free region for different
heights in the complex plane. There is no need for zero-free regions for small (s),
however, as the Riemann hypothesis has been verified up to a height H. The most
recent computation of H is from Platt and Trudgian [31].

Lemma 3.1 ([31]). If ζ(β + it) = 0 for any |t| ≤ 3 · 1012 then β = 12 .

The most recent explicit version of the “classical” zero-free region is from Moss-
inghoff and Trudgian [26].

Lemma 3.2 ([26]). For |t| ≥ 2 there are no zeros of ζ(β + it) in the region
β ≥ 1 − ν1 (t), where
1
ν1 (t) =
R0 log |t|
anda R0 = 5.5666305.

a Thisvalue of R0 is lower than that appearing in [26, Theorem 1]. However, since the Riemann
hypothesis has now been verified to a higher height (Lemma 3.1), we can take R0 = 5.5666305 as
discussed in [26, Sec. 6.1].
 June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

1212 M. Cully-Hugill & D. R. Johnston

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

Ford [16] gave a version of the classical result which improves on Mossinghoff
and Trudgian’s result for large |t|.

Lemma 3.3. For |t| ≥ 5.45 · 108 there are no zeros of ζ(β + it) in the region
β ≥ 1 − ν2 (t), where
1
ν2 (t) = ,
R(|t|) log |t|
with
J(t) + 0.685 + 0.155 log log t
R(t) =  ,
0.0196
log t 0.04962 −
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

J(t) + 1.15
and
1
J(t) = log t + log log t + log(0.63).
6

Proof. This result is almost identical to [16, Theorem 3], except we use an improved
constant in J(t). Ford’s (1.6) can be replaced with a more recent result due to Hiary
[18, Theorem 1.1]. We note however, that an error was recently discovered in [18]
(see [29, §2.2.1]). Despite this, Hiary’s result can be recovered (and in fact improved)
as discussed in the preprint [19].

For very large t, we use an explicit Vinogradov–Korobov zero-free region, also

due to Ford [16].
Lemma 3.4 ([16, Theorem 5]). For |t| ≥ 3 there are no zeros of ζ(β + it) in the
region β ≥ 1 − ν3 (t) where
1
ν3 (t) = 2/3
(3.1)
c log |t|(log log |t|)1/3
and c = 57.54.
To use the widest zero-free region we set
⎧
⎪ 1
⎨ , if |t| ≤ 3 · 1012 ,
ν(t) = 2 (3.2)
⎪
⎩
max{ν1 (t), ν2 (t), ν3 (t)}, otherwise,
noting that ν2 (t) ≥ ν1 (t) for t ≥ exp(46.3) and ν3 (t) ≥ ν2 (t) for t ≥ exp(54599).

4. Wolke’s Method
In this section, we prove an explicit bound for the integral in Theorem 2.1,
κ+iT
1 xs
F (s) ds,
2πi κ−iT s
June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

The error term in the explicit formula of Riemann–von Mangoldt 1213

Λ(n)n−s = −(ζ  /ζ)(s).

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

for the case κ = 1 + 1/ log x and F (s) = n≥1

Theorem 4.1. Let α ∈ (0, 1] and ω ∈ [0, 1]. There exists constants K and xK such
α
that if x ≥ xK and max{51, log x} < T < x2 − 1,
1+ε+iT   s  xρ  
1 ζ x Kx
− (s) ds = x − + O∗ (log x)1−ω , (4.1)
2πi 1+ε−iT ζ s ρ T
ρ=β+iγ
|γ|≤T

where ε = 1/ log x. Some corresponding values of α, ω, K and xK are given in

Table A.1 in the Appendix.
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

Although setting ω = 1 gives the best result asymptotically, the resulting con-
stant will be quite large unless xK is very large. For the best results we take ω closer
to 1 for large x and small T , and ω closer to 0 for small x and large T .
To prove Theorem 4.1 we make [38, Theorem 2] explicit. The restriction T > 51 is
so we can directly use some results from [13], which contains a proof of an expression
similar to (4.1) albeit with a weaker error term. The restriction T > log x is not
necessary but allows us to obtain a better value of K for large x.
In what follows, let ω ≥ 0 be a parameter to be optimized in our computations.
The value ω appearing in Theorem 4.1 will be given by
⎧
⎨ω, if ω < 1,
ω=
⎩1, if ω ≥ 1.

We also make use of explicit zero-free regions, and set ν(t) as in (3.2).
We now prove a couple of preliminary lemmas. In what follows, N (T ) denotes
the number of zeros of ζ(s) with imaginary part up to height T .

Lemma 4.2. Let t > 1. Then N (t + 1) − N (t − 1) < log t.

Proof. Dudek [13, Lemma 2.6] proves this lemma for t > 50. For 1 < t ≤ 50 we
used Odlyzko’s tables [27] to verify the lemma manually.

Lemma 4.3. Let T > 51 and x > e. There exists a τ ∈ (T − 1/(log x)ω , T +
1/(log x)ω ) such that when s = σ + iτ with σ > −1, we have
  
ζ 
 (s) < (log x)ω (log2 T + log T ) + 20 log T.
ζ 

Proof. By Lemma 4.2,

N (T + 1/(log x)ω ) − N (T − 1/(log x)ω ) ≤ N (T + 1) − N (T − 1) < log T.
(4.2)
As there are at most log T zeros of ζ(s) with imaginary part in the interval (T −
1/(log x)ω , T + 1/(log x)ω ), we can split (T − 1/(log x)ω , T + 1/(log x)ω ) into at most
log T + 1 zero-free regions. At least one of these regions will have height greater
 June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

1214 M. Cully-Hugill & D. R. Johnston

than
by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

2
.
(log x)ω (log T + 1)
We define τ to be the midpoint of such a region so that
1
|τ − γ| ≥ (4.3)
(log x)ω (log T + 1)
for all zeros ρ = β + iγ. Next we use the following result which holds for all σ > −1
and τ > 50 [13, p. 187]
ζ  1
(s) = + O∗ (19 log τ ). (4.4)
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

ζ s−ρ
|γ−τ |<1

Using (4.2) and (4.3),

 
   
 1  1
 ≤ < (log x)ω (log T + 1) log T
 
|γ−τ |<1 s − ρ  |γ−τ |<1 |τ − γ|
which completes the proof of the lemma upon noting that 19 log τ ≤ 20 log T .

We are now in a position to prove Theorem 4.1.

Proof of Theorem 4.1. Letting τ be as in Lemma 4.3, we define the contour

C = C1 ∪ C2 ∪ C3 ∪ C4 , where
C1 = [1 + ε − iτ, 1 + ε + iτ ], C2 = [1 + ε + iτ, −1 + iτ ],
C3 = [−1 + iτ, −1 − iτ ], C4 = [−1 − iτ, 1 + ε − iτ ].
Cauchy’s residue theorem gives
1+ε+iτ   s  xρ
1 ζ x
− (s) ds = x − − log(2π)
2πi 1+ε−iτ ζ s ρ
|γ|≤τ
   s
1 ζ x
− − (s) ds, (4.5)
2πi C2 ∪C3 ∪C4 ζ s
noting that (ζ  /ζ)(0) = log(2π). We begin by converting the left-hand side of (4.5)
into an integral involving T as opposed to τ . To do this, we note that when (s) =
1 + ε, the main theorem in [12] gives
     
ζ   
 (s) ≤  ζ (1 + ε) < log x.
ζ  ζ 
Hence,
    s 
 1 1+ε+iτ
ζ x  ex
 − (s) ds ≤ |τ − T | log x
 2πi ζ s  2π(T − 1)
1+ε+iT
ex
≤ (log x)1−ω (4.6)
2π(T − 1)
June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

The error term in the explicit formula of Riemann–von Mangoldt 1215

and identically for the integral from 1 + ε − iτ to 1 + ε − iT . Therefore,

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

1+ε+iτ   s 1+ε+iT   s
1 ζ x 1 ζ x
− (s) ds = − (s) ds
2πi 1+ε−iτ ζ s 2πi 1+ε−iT ζ s
 
ex
+ O∗ (log x)1−ω .
π(T − 1)
Next we consider the sum on the right-hand side of (4.5). In the case T < τ ,
 xρ  xρ  xρ
= + .
ρ ρ ρ
|γ|≤τ |γ|≤T T <|γ|≤τ
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

By (4.2) we then have

 
 
  xρ  1−ν(τ )
  ≤ 2 log τ x ≤
2x log(T + 1)
.
 ρ  T xν(T +1) T
T <|γ|≤τ 

The case T > τ is similar, with an error bounded by 2x log T /xν(T ) (T − 1).
Hence,
 xρ  xρ  
2x log(T + 1)
= + O∗ ν(T +1) . (4.7)
ρ ρ x (T − 1)
|γ|≤τ |γ|≤T

Finally we deal with the integral on the right-hand side of (4.5). For the integral
over C3 we use the inequality [13, Lemma 2.3]
  
ζ 
 (s) < 9 + log |s|,
ζ 

which holds when (s) = −1. In particular,

    s  √
 1 ζ x  2
 − (s) ds  ≤ 2τ 9 + log( τ + 1)
 2πi ζ s  2π x
C3

(T + 1)(9 + log( (T + 1)2 + 1)
≤ .
πx
Next we bound the integral over C2 . The same bound will also hold for the integral
over C4 by symmetry.
Case 1: ω = 0.
When ω = ω = 0 we bound the integral over C2 using Lemma 4.3. That is,
    s 
 1 ζ x  log2 T + 21 log T 1+ε σ
 − (s) ds ≤ x dσ
 2πi ζ s 2πτ
C2 −1

ex(log2 T + 21 log T )
≤ . (4.8)
2π log x(T − 1)
 June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

1216 M. Cully-Hugill & D. R. Johnston

Case 2: ω > 0.
by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

In the case where ω > 0 we need to be more careful since the expression in (4.8) is
not O(x/T ) (unless further restrictions are placed on T ). To overcome this issue, we
let D ∈ (0, 1) be a parameter that we will optimize for different entries in Table A.1
and let σ0 = σ0 (x, α) := 1 − Dν(xα ). We then split C2 into C2 = C21 ∪ C22 where

C21 = [1 + ε + iτ, σ0 + iτ ], C22 = [σ0 + iτ, −1 + iτ ].

By Lemma 4.3
    s 
 1 ζ x  (log x)ω (log2 T + log T ) + 20 log T σ0
 − (s) ds ≤ xσ dσ
 2πi ζ s  2πτ
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

C22 −1

(log x)ω−1 (log2 T + log T ) + 20 log T / log x

≤x .
2πxDν(xα ) (T − 1)
(4.9)

For the integral over C21 we follow Wolke and use the following formula for ζ  /ζ
[35, Lemma 2]. Let, for y = x1/4 > 1,
⎧
⎪
⎨Λ(n), 1 ≤ n ≤ y,
Λy (n) = 2
⎪
⎩Λ(n) log(y /n) , y ≤ n ≤ y 2 .
log y
Then if s = 1 and s is not a zero of ζ,
 Λy (n) y 1−s − y 2(1−s) ∞
ζ 1  y −2q−s − y −2(2q+s)
− (s) = + −
ζ 2
ns (1 − s)2 log y log y q=1 (2q + s)2
n≤y

1  y ρ−s − y 2(ρ−s)
−
log y ρ (s − ρ)2

= Z1 (s) + Z2 (s) + Z3 (s) + Z4 (s), say.

For Z1 ,
 
 1
 xs  1  1+ε
x σ
 2πi Z1 (s) ds ≤ Λy (n) dσ
C21 s 2πτ 2 σ0 n
n≤y

1  (x/n)1+ε
≤ Λy (n)
2π(T − 1) 2
log(x/n)
n≤y

ex  Λ(n)
≤ c log(x/n)
2π(T − 1) 1/2
n
n≤x

ex  Λ(n)
≤ . (4.10)
π log x(T − 1) 1/2
nc
n≤x
June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

The error term in the explicit formula of Riemann–von Mangoldt 1217

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

Then by the corollary of the main theorem in [33],

 Λ(n) 1 8 1
≤ log x − γ + 2 ≤ log x, (4.11)
1/2
n 2 log x 2
n≤x

for x > 50. By partial summation, we then deduce from (4.11) that
 Λ(n) 1
c
≤ log x.
1/2
n e
n≥x

This gives an O(x/T ) bound for (4.10).

For Z2 ,
 
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

 1 xs  1 1+ε
 1−σ 1−σ
)xσ dσ
Z (s) ds  ≤ 2πτ 3 log y (x +x
4 2
 2πi 2
s
C21 σ0
 1+3(1+ε) 1+(1+ε)

2 4x 4 x 2
≤ +2
π(T − 1)3 log x 3 log x log x
 
2 3/4
4x e + e1/2
3
≤ .
π(T − 1)3 log2 x
For Z3 ,
   ∞ ∞

 1 xs  1 1+ε  
 Z3 (s) ds ≤ y −σ y −2q + y −2σ y −4q xσ dσ
 2πi s 2πτ 3 log y
C21 σ0 q=1 q=1
 
2 (ex)3/4 1 (ex)1/2 1
≤ +
π(T − 1)3 log x 2
log(x ) y − 1 log(x ) y − 1
3/4 1/2 4

 
2x 4 e3/4 2e1/2
= + .
π(T − 1)3 log2 x 3 x3/4 − x1/4 x3/2 − x1/2
For Z4 , we first note that
1
1 + ε − σ0 = + Dν(xα ) = dα (x)ν(xα ), (4.12)
log x
where dα (x) := D + 1/ν(xα ) log x is non-increasing. Moreover, for a zero ρ = β + iγ
with |γ| ≤ 2T + 2 < xα ,
σ0 − β > (1 − D)ν(xα ) =: να (x). (4.13)
Now,
 
 1 xs 
 Z4 (s) ds (4.14)
 2πi s
C21

1  ex 1+ε
y β−σ + y 2(β−σ)
≤ dσ
2π log y ρ τ σ0 |σ + iτ − ρ|2
|γ|≤2T +2
 June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

1218 M. Cully-Hugill & D. R. Johnston

 σ−1  σ−1
by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

x β−1 x
 1+ε y + y 2(β−1)
1 x y y2
+ dσ
2π log y ρ τ σ0 |σ + iτ − ρ|2
|γ|>2T +2

2x
= (eF1 (x, τ ) + F2 (x, τ )), say. (4.15)
πτ log x
To bound F1 (x, τ ) we will use the estimate
1+ε
dσ 1
≤ dα (x)v(xα ). (4.16)
σ0 |σ + iτ − ρ|2 |τ − γ|2
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

whenever |τ − γ| > 2. On the other hand, when |τ − γ| ≤ 2, we use the estimate

1+ε 1+ε
dσ dσ 1 1
≤ = −
σ0 |σ + iτ − ρ|2 σ0 (σ − β)2 σ0 − β 1+ε−β
dα (x)
< . (4.17)
(dα (x) + 1 − D)vα (x)
Using (4.16), (4.17) and Lemma 4.2 we have that F1 (x, τ ) is bounded above by
⎛ ⎞
 ∞
 
1 ⎜ ⎟ 1+ε
dσ
⎜ + ⎟
ν
xα (x)/4 ν
+x α (x)/2 ⎝ ⎠ |σ + iτ − ρ|2
|γ|≤2T +2 h=2 |γ|≤2T +2 σ0
|τ −γ|≤2 2(h−1)<|τ −γ|≤2h

1 2dα (x) log τ
≤ ν (x)/4
xα + xνα (x)/2 (dα (x) + 1 − D)vα (x)
∞

 1
α
+ 2dα (x)ν(x ) log(2T + 1)
m=1
(2m)2
 
2dα (x) log(xα ) 1 v(xα )π 2
≤ ν (x)/4 + . (4.18)
xα + xνα (x)/2 (dα (x) + 1 − D)vα (x) 24
Then,
 1 1+ε
F2 (x, τ ) ≤  γ 2 (x3/4 )σ−1 + (x1/2 )σ−1 dσ
  σ0
ρ  
|γ|>2T +2 2
 
e3/4 e1/2 1
≤4 + 2
log(x3/4 ) log(x1/2 ) ρ
|γ|
 
4 4 3/4 1/2 1
= e + 2e 2
,
log x 3 ρ
|γ|

where ρ 1/|γ|2 ≤ 0.04621 by [3, Example 1].

June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

The error term in the explicit formula of Riemann–von Mangoldt 1219

Combining all of our estimates we see that

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

1+ε+iT   s  xρ
1 ζ x
− (s) ds = x − + E(x, T ),
2πi 1+ε−iT ζ s ρ
|γ|≤T

where |E(x, T )| is bounded above by


ex 2x log(T + 1) (T + 1)(9 + log( (T + 1)2 + 1))
log(2π) + log x + +
π(T − 1) xν(T +1) T πx
ex(log2 T + 21 log T )
+ (4.19)
π log x(T − 1)
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

when ω = ω = 0 and

ex 1−ω 2x log(T + 1) (T + 1)(9 + log( (T + 1)2 + 1))
log(2π) + (log x) + +
π(T − 1) xν(T +1) T πx

(log x)ω−1 (log2 T + log T ) + 20 log T / log x x
+2 x +
2πx Dν(x α)
(T − 1) π(T − 1)
 
2 3/4 1/2
4x e +e  
3 2x 4 e3/4 2e1/2
+ + +
π(T − 1)3 log2 x π(T − 1)3 log2 x 3 x3/4 − x1/4 x3/2 − x1/2
 
4αexdα (x) 1 v(xα )π 2
+ +
π(T − 1)(xνα (x)/4 + xνα (x)/2 ) (dα (x) + 1 − D)vα (x) 24
  
8x 4 3/4 1/2 1
+ e + 2e (4.20)
π(T − 1) log2 x 3 ρ
|γ|2

otherwise. Note that the terms multiplied by 2 in (4.20) are those that occur
when bounding both the integral over C2 and C4 . Thus, we can write E(x, T ) =
O∗ (Kx(log x)1−ω /T ) where K can be computed by evaluating an upper bound for
each error term in (4.19) or (4.20) in the range x ≥ xK .

Extra care must be taken when bounding some of the error terms in the above
proof due to ν(t) being defined as a composite function. Most notably, we need
to be careful of the behavior at the crossover point λ = 54598.16 . . . such that
ν3 (t) ≥ ν2 (t) for all t ≥ exp(λ) or equivalently ν3 (tα ) ≥ ν2 (tα ) for all t ≥ exp(λ/α).
For example, the error in (4.9) can be written as H(x, T )(x(log x)1−ω /T ) where
H decreases in x whilst ν(T ) = 1/2, and increases in x whilst ν(T ) = ν1 (T ) or
ν(T ) = ν2 (T ). Thus, in most cases, an upper bound for H(x, T ) occurs at or beyond
α
x = exp(λ/α) corresponding to the upper bound T < x 4−2 < xα for T . As a result,
in the cases where xK < exp(λ/α) we evaluated (4.9) at x = exp(λ/α) to obtain
an upper bound. In such cases we also had to verify that (4.9) was decreasing for
x ≥ exp(λ/α). A similar treatment is required for the terms in (4.7) and (4.18).
 June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

1220 M. Cully-Hugill & D. R. Johnston

We also remark that the error term

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

2x
π(T − 1)
in (4.20) is bounded below by 2x/πT independent of the choice of α or ω. Thus, K
is bounded below by 2/π = 0.63661 . . . when ω = 1.
Since our computations rely heavily on zero-free regions, we can obtain signif-
icant improvements by assuming the Riemann hypothesis. In particular, we have
the following result.

Theorem 4.4. Assuming the Riemann hypothesis, we can take K = 0.6373 in (4.1)
for ω = 1, α = 1/2 and log x ≥ 1000.
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

Proof. Redefine ν(t) = 12 in the proof of Theorem 4.1. The parameters we used to
obtain K = 0.6373 were D = 0.4 and ω = 3.

Remark. With more work, the value of K in Theorem 4.4 can be lowered and in
fact made arbitrarily small as x → ∞. In particular, one could bound the integral
over C2 using conditional estimates on ζ  /ζ (e.g. [36, Corollary 1]).

5. Proof of Theorem 1.2

Using the results of Secs. 2 and 4, we now prove Theorem 1.2. To begin with, we
α
let x ≥ xM ≥ e40 , max{51, log x} < T < x2 − 1, κ = 1 + 1/ log x and set an = Λ(n)
in Theorem 2.1 to obtain
κ+iT   s
1 ζ x
ψ(x) = − (s) ds
2πi κ−iT ζ s
⎛ ⎞
κ  λ  κu
x Λ(n) 1 e
+ O∗ ⎝ + Λ(n) 2 du⎠. (5.1)
πλT nκ πT θ /T u
n≥1 | log(x/n)|≤u

For the first term, we have by Theorem 4.1,

κ+iT   s  xρ
1 ζ x x
− (s) ds = x + + O∗ K (log x)1−ω , (5.2)
2πi κ−iT ζ s ρ T
|γ|≤T

for some constant K corresponding to xK = xM . Then, the first sum in (5.1) is

bounded by
xκ  Λ(n) ex log x
≤ (5.3)
λπT nκ λπT
n≥1

by the main theorem in [12]. For the second term in (5.1), the condition | log(x/n)| ≤
u is equivalent to xe−u ≤ n ≤ xeu . For all u ∈ [0, λ] we have e−u ≥ 1 − u and
eu ≤ 1 + (c0 − 1)u with
eλ − 1
c0 = c0 (λ) = + 1.
λ
June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

The error term in the explicit formula of Riemann–von Mangoldt 1221

Hence, the sum simplifies to

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

1 λ  eκu λ
eκλ 1
Λ(n) du ≤ πT θ  /T u2 I(x,u) Λ(n)du, (5.4)
πT θ  /T | log(x/n)|≤u u2

where
I(x, u) = {n ≥ 1 : x − ux ≤ n ≤ x + (c0 − 1)ux}.
Let x− = max{x − ux, 0} and x+ = x + (c0 − 1)ux for θ /T ≤ u ≤ λ. Defining

θ(x) = log p,
p≤x
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

we have by an explicit form of the Brun–Titchmarsh theorem [25, Theorem 2],

2 log x+
θ(x+ ) − θ(x− ) ≤ (x+ − x− )
log(x+ − x− )
2 log(x + (c0 − 1)ux)
≤ c0 ux
log(c0 ux)
≤ c0 ux · E1 (x, T ),
where
2 log(x + (c0 − 1)λx)
E1 (x, T ) = .
log(c0 θ x/T )
To obtain the corresponding inequality for ψ, we use [4, Corollary 5.1], which states
that for all x ≥ e40 ,
0 < ψ(x) − θ(x) < a1 x1/2 + a2 x1/3 ,
with a1 = 1 + 1.93378 · 10−8 and a2 = 1.0432. Thus,

Λ(n) ≤ ψ(x+ ) − ψ(x− ) + log x
n∈I(x,u)
√ √
< c0 ux · E1 (x, T ) + a1 x+ + a2 3 x+ + log x
≤ c0 ux · E1 (x, T ) + E2 (x),
where
E2 (x) = a1 (x + (c0 − 1)λx)1/2 + a2 (x + (c0 − 1)λx)1/3 + log x.
Substituting this into (5.4), we obtain
 
eκλ λ
1  eκλ λ
c0 x · E1 (x, T ) E2 (x)
2
Λ(n)du ≤ + du
πT θ  /T u πT θ  /T u u2
I(x,u)
   
eκλ c0 x · E1 (x, T ) λ E2 (x)
≤ log  T + . (5.5)
π T θ θ
 June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

1222 M. Cully-Hugill & D. R. Johnston

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

Combining (5.2), (5.3) and (5.5), the proof of Theorem 1.2 is complete upon opti-
mizing over λ to compute values of xM , α ∈ (0, 1/2] and M such that
 xρ  
M x log x
ψ(x) = x − + O∗ ,
ρ T
ρ=β+iγ
|γ|≤T

α
for all x ≥ xM and max{51, log x} < T < x2 − 1. Note we cannot take α > 1/2, as
√
E2 (x) would not be O(x log x/T ) for values of T asymptotically larger than x.
Under assumption of the Riemann hypothesis, we also have the following.
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

Theorem 5.1. Assuming the Riemann hypothesis, we can take M = 4.150 in (1.5)
with log x ≥ 1000 and α = 1/2.

Proof. Calculate M as above using λ = 0.52 and the value of K = 0.6373 from
Theorem 4.4.

6. Application: Primes Between Consecutive Powers

For sufficiently large x, the best unconditional short-interval result for primes is
√
[x, x + x0.525 ], from Baker et al. [1]. This can be narrowed to (x, x + C x log x]
for some constant C if assuming the Riemann hypothesis, as per Cramér’s result
[6]. The latest explicit version of this is [5, Theorem 1.5], with C = 22/25 for all
x ≥ 4. Legendre conjectured that something just better than this should be true:
that there should be a prime between n2 and (n + 1)2 for all positive integers n.
√
This is approximately equivalent to primes in intervals of the form (x, x + 2 x].
Although proving Legendre’s conjecture is out of reach even under RH, we do know
there are primes between higher consecutive powers. Using Ingham’s method from
[20], it was shown in [7] that there are primes between consecutive cubes, n3 and
(n + 1)3 , for all n ≥ exp(exp(32.892)).
It is also possible to find primes between higher consecutive powers for all pos-
itive n. In [7], this is done using an explicit version of Goldston’s estimate for the
error in the prime number theorem [17]. We can now use Theorem 1.2 in place of
Goldston’s result, and prove Theorem 1.3.
1
To consider the interval (nm , (n + 1)m ), we set n = x m and look at the slightly
smaller interval (x, x + h] with h = mx1−1/m . Theorem 1.2 implies
 
  (x + h)ρ − xρ 
ψ(x + h) − ψ(x) ≥ h −   − M G(x, h) (6.1)
 ρ  T
ρ=β+iγ
|γ|≤T

for all x ≥ xM and max{51, log x} < T < (xα −2)/2, where G(x, h) = (x+h) log(x+
h) + x log x, and values for xM , α and M are given in Table A.2. In order to find
at least one prime in (x, x + h], (6.1) needs to be positive. Fixing an m, we want
to maximize (6.1) with respect to T , and solve for x. This process follows the same
June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

The error term in the explicit formula of Riemann–von Mangoldt 1223

steps as in [7, Sec. 4], and we set T = xμ for some μ ∈ (0, 1).b We then arrive at
by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

a similar condition in [7, Eq. (21)]: that there are primes between consecutive mth
powers for all x satisfying
G(x, h) E(x)
1 − F (x) − M + >0 (6.2)
xμ h h
where F (x) and E(x) are defined as in Eqs. (19) and (20), respectively, of [7]. A
zero-free region is used in this definition of F (x), so we use the ν(t) defined in Sec. 3
of this paper, instead of that used in [7].
For log xM = 103 we can take M = 0.6651, so for m = 140 we can take μ =
0.0080155 to have (6.2) hold for all log x ≥ 4242. Note that with these values
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

the condition on T in Theorem 1.2 is satisfied for all log x ≥ 375. The interval
estimates for primes in [8] (corrected from the original paper [9]) for x ≥ 4 · 1018
and x ≥ e600 verify that there are primes between consecutive 140th powers for
log(4 · 1018 ) ≤ log x ≤ 4367. The calculations of [28] verify the remaining smaller
values of x.

7. Application: The Error Term in the Prime Number Theorem

The prime number theorem is equivalent to the statement ψ(x) ∼ x. For large x,
the best unconditional estimates on the error |ψ(x)−x| are from Platt and Trudgian
[32, Theorem 1]. For R = 5.5734125 they give values for X, A, B and C such that
   B   
 ψ(x) − x  log x log x
 ≤A exp −C (7.1)
 x  R R

for all log x ≥ X. To obtain (7.1), Platt and Trudgian employ a method of Pintz [30]
and use Dudek’s error term for the Riemann–von Mangoldt formula (Theorem 1.1).
In this section, we prove Theorem 1.4, which is a non-trivial improvement on
Platt and Trudgian’s result. Most of the improvement comes from using Theorem 1.2
in place of Dudek’s error term. We will also incorporate some other recent results.
Firstly, we use the smaller value R = 5.5666305, which is the same R0 appearing
in the zero-free region in Lemma 3.2. In the following lemma, we also make small
improvements to some of the zero-density estimates in [23].

Lemma 7.1. Let N (σ, T ) be the number of zeros ρ = β + iγ of the Riemann

zeta-function with σ < β ≤ 1 and 0 ≤ γ ≤ T . Then, for the values of C1 (σ) and
C2 (σ) in Table 2, we have
N (σ, T ) ≤ C1 (σ)T 8(1−σ)/3 (log T )5−2σ + C2 (σ) log2 T.

Proof. Using Platt and Trudgian’s verification of the Riemann hypothesis up to

H0 = 3 · 1012 [31], and the divisor function estimate in Theorem 2 of [10] to replace

b Note that μ here is denoted α in [7].

 June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

1224 M. Cully-Hugill & D. R. Johnston

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

Table 2. Some values of C1 (σ) and C2 (σ) for

Lemma 7.1.
σ α d C1 (σ) C2 (σ)
0.980 0.063 0.336 15.743 2.214
0.982 0.063 0.336 15.878 2.204
0.984 0.061 0.336 16.013 2.187
0.986 0.061 0.336 16.148 2.171
0.988 0.060 0.337 16.284 2.148
0.990 0.060 0.337 16.421 2.132
0.992 0.058 0.337 16.558 2.115
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

(3.13) of [23], we can recalculate the constants in [23, Lemma 4.14]. Using the
notation from [23], we want to optimize over k, μ, α, δ, d, H, and η. We chose
H = H0 − 1, η = 0.2535, k = 1, μ = 1.237, δ = 0.313 and optimized over the other
parameters for each σ.

Proof of Theorem 1.4. We only require a slight modification of the proof of

[32, Theorem 1]. First we assume log x ≥ 1000 so that we may take xM = exp(1000),
α = 1/10 and M = 2.045 in Theorem 1.2. The argument in [32, pp. 874–875] then
follows through with minimal modification so that if C1 (σ) and C2 (σ) are as in
Lemma 7.1 and
⎡     5−2σ ⎤−1
10 − 16σ log x0 log x0
k(σ, x0 ) = ⎣exp ⎦ ,
3 R R
  
log x0
C3 (σ, x0 ) = M (log x0 ) exp −2 k(σ, x0 ),
R
 
2 log x0
C4 (σ, x0 ) = xσ−1
0 + 1.8642 k(σ, x0 ),
π R
  
log x0 log x0
C5 (σ, x0 ) = 8.01 · C2 (σ) exp −2 k(σ, x0 ),
R R

A(σ, x0 ) = 2.0025 · 25−2σ · C1 (σ) + C3 (σ, x0 ) + C4 (σ, x0 ) + C5 (σ, x0 ),

then
    5−2σ   
 ψ(x) − x  2
10 − 16σ log x
  ≤ A(σ, x0 ) log x exp
 x  R 3 R
for all σ ∈ [0.75, 1) and x > x0 provided A(σ, x) is decreasing for x > x0 . The
only difference between our formulae and those in [32, p. 875] is the expression for
C3 (σ, x0 ). Taking x0 = exp(X) it is then possible to compute the values of A in
Table 1. The values of σ in Table 1 were chosen to optimize the value of 0 .
June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

The error term in the explicit formula of Riemann–von Mangoldt 1225

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

Remark. Using Theorem 1.2 meant that the C3 (σ, x0 ) term in the above proof
became essentially negligible for the values of x0 considered. Thus, reducing M is
unlikely to substantially improve Theorem 1.4 using the above method.
We also have the following corollary (cf. [32, Corollary 1]).
Corollary 7.2. For each row {X, A, B, C} in Table 1 we have
   B   
 θ(x) − x 
  ≤ A1 log x exp −C
log x
, for all log x ≥ X,
 x  R R
where A1 = A + 0.1.
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

Proof. By [4, Corollary 5.1], we have for x ≥ exp(1000) that

ψ(x) − θ(x) < a1 x1/2 + a2 x1/3 ,
with a1 = 1 + 1.99986 · 10−12 and a2 = 1 + 1.936 · 10−8 . The result then follows by
a straightforward computation since
   
 θ(x) − x  ψ(x) − θ(x)  ψ(x) − x 
 ≤ +  .
 x  x x 

Corollary 7.2 allows us to improve on existing knowledge of an inequality due

to Ramanujan. Namely, in one of his notebooks, Ramanujan proved that
ex x
π(x)2 < π (7.2)
log x e
holds for sufficiently large x [2, pp. 112–114].
It is still an open problem to determine the largest value of x for which (7.2)
holds. However, it is widely believed that the last integer counterexample occurs
at x = 38, 358, 837, 682. In fact, this follows under assumption of the Riemann
hypothesis [14, Theorem 1.3].
Using their expression for |θ(x) − x|, Platt and Trudgian [32, Theorem 2] were
able to show (unconditionally) that (7.2) holds for x ≥ exp(3915). Substituting our
results for Corollary 7.2 when X = 3600 into the formulae on [32, p. 879] gives a
small improvement. In particular, we get that Ramanujan’s inequality (7.2) holds
for x ≥ exp(3604) (Corollary 1.5).
We also note that the second author recently showed that Ramanujan’s inequal-
ity (7.2) holds for 38, 358, 837, 683 ≤ x ≤ exp(103) [21, Theorem 5.1]. However, we
are not able to improve on this result here as Theorem 1.4 and Corollary 7.2 only
give good bounds on prime counting functions for large x ≥ exp(1000).

Acknowledgments
Thanks to all our colleagues at UNSW Canberra. Particularly our supervisor Tim
Trudgian for his constant support, Aleks Simonič for helping us with those pesky
semicircles, and Ethan Lee for the long discussions over tea. We also thank Ramaré
for his insights and correspondence.
 June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

1226 M. Cully-Hugill & D. R. Johnston

Appendix A: Tables A.1 and A.2

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

Table A.1. Some corresponding values of xK , α, ω,

D and K for Theorem 4.1.

log(xK ) α ω ω D K
40 1/2 0 0 — 2.053
103 1/2 0 0 — 1.673
1010 1/2 0.3 0.3 0.54 3.191
1013 1/2 1 1.4 0.50 0.6367
103 1/10 0.2 0.2 0.45 2.596
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

1010 1/10 0.9 0.9 0.54 11.77

103 1/100 0.8 0.8 0.52 2.186
1010 1/100 1 1.5 0.50 0.6367

Table A.2. Some corresponding values of xM ,

α, λ and M for Theorem 1.2.

log(xM ) α λ M
40 1/2 0.48 6.431
103 1/2 0.52 5.823
1010 1/2 0.52 4.143
1013 1/2 0.52 4.140
103 1/10 1.05 2.045
1010 1/10 1.06 1.384
103 1/100 1.80 0.6651
1010 1/100 1.88 0.6269

Note that although we could have considered larger values of α ∈ (0, 1] in

Table A.1, the restriction α ≤ 1/2 is required for Theorem 1.2.

References
[1] R. C. Baker, G. Harman and J. Pintz, The difference between consecutive primes. II,
Proc. London Math. Soc. 83(3) (2001) 532–562.
[2] B. C. Berndt, Ramanujan’s Notebooks: Part IV (Springer-Verlag, 1994).
[3] R. Brent, D. Platt and T. Trudgian, Accurate estimation of sums over zeros of the
Riemann zeta-function, Math. Comp. 90 (2021) 2923–2935.
[4] S. Broadbent, H. Kadiri, A. Lumley, N. Ng and K. Wilk, Sharper bounds for the
Chebyshev function θ(x), Math. Comput. 90(331) (2021) 2281–2315.
[5] E. Carneiro, M. B. Milinovich and K. Soundararajan, Fourier optimization and prime
gaps, Comment. Math. Helv. 94(3) (2019) 533–568.
June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

The error term in the explicit formula of Riemann–von Mangoldt 1227

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

[6] H. Cramér, On the order of magnitude of the difference between consecutive prime
numbers, Acta Arith. 2 (1936) 23–46.
[7] M. Cully-Hugill, Primes between consecutive powers, J. Number Theory 247 (2023)
100–117.
[8] M. Cully-Hugill and E. S. Lee, Explicit interval estimates for prime numbers, preprint
(2022), arXiv:2103.05986.
[9] M. Cully-Hugill and E. S. Lee, Explicit interval estimates for prime numbers, Math.
Comput. 91(336) (2022) 1955–1970.
[10] M. Cully-Hugill and T. Trudgian, Two explicit divisor sums, Ramanujan J. 56 (2021)
141–149.
[11] H. Davenport, Multiplicative Number Theory, Vol. 74, 2nd edn. (Springer-Verlag, New
York, 1980).
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

[12] H. Delange, Une remarque sur la dérivée logarithmique de la fonction zêta de Rie-
mann, Colloq. Math. 53(2) (1987) 333–335.
[13] A. W. Dudek, An explicit result for primes between cubes, Funct. Approx. Comment.
Math. 55(2) (2016) 177–197.
[14] A. W. Dudek and D. J. Platt, On solving a curious inequality of Ramanujan, Exp.
Math. 24(3) (2015) 289–294.
[15] A. Fiori, H Kadiri and J. Swidinsky, Sharper bounds for the error term in the prime
number theorem, preprint (2022), arXiv:2206.12557.
[16] K. Ford, Zero-free regions for the Riemann zeta function, in Number Theory for the
Millennium, II (A K Peters, Natick, MA, 2002), pp. 25–56.
[17] D. A. Goldston, On a result of Littlewood concerning prime numbers. II, Acta. Arith.
43(1) (1983) 49–51.
[18] G. A. Hiary, An explicit van der corput estimate for ζ(1/2 + it), Indag. Math. 27(2)
(2016) 524–533.
[19] G. A Hiary, D. Patel and A. Yang, An improved explicit estimate for ζ(1/2 + it),
preprint (2022), arXiv:2207.02366.
[20] A. E. Ingham, On the difference between consecutive primes, Q. J. Math. 8(1) (1937)
255–266.
[21] D. R. Johnston, Improving bounds on prime counting functions by partial verification
of the Riemann hypothesis, Ramanujan J. (2022).
[22] D. R. Johnston and A. Yang, Some explicit estimates for the error term in the prime
number theorem, preprint (2022), arXiv:2204.01980.
[23] H. Kadiri, A. Lumley and N. Ng, Explicit zero density for the Riemann zeta function,
J. Math. Anal. Appl. 465(1) (2018) 22–46.
[24] H. Montgomery and R. Vaughan, Multiplicative Number Theory I: Classical Theory
(Cambridge University Press, Cambridge, 2007).
[25] H. L. Montgomery and R. C. Vaughan, The large sieve, Mathematika 20 (1973)
119–134.
[26] M. J. Mossinghoff and T. S. Trudgian, Nonnegative trigonometric polynomials and a
zero-free region for the Riemann zeta-function, J. Number Theory 157 (2015) 329–
349.
[27] A. Odlyzko, Tables of zeros of the Riemann zeta function (2021),
http://www.dtc.umn.edu/õdlyzko/zeta tables/index.html.
[28] T. Oliveira e Silva, S. Herzog and S. Pardi, Empirical verification of the even Goldbach
conjecture and computation of prime gaps up to 4 · 1018 , Math. Comput. 83(288)
(2014) 2033–2060.
[29] D. Patel, Explicit sub-Weyl bound for the Riemann zeta function. PhD thesis, The
Ohio State University (2021).
 June 16, 2023 9:20 WSPC/S1793-0421 203-IJNT 2350059

1228 M. Cully-Hugill & D. R. Johnston

by NATIONAL AUTONOMOUS UNIVERSITY OF MEXICO (UNAM) on 08/06/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

[30] J. Pintz, On the remainder term of the prime number formula. II. On a theorem of
Ingham, Acta Arith. 37 (1980) 209–220.
[31] D. Platt and T. Trudgian, The Riemann hypothesis is true up to 3 · 1012 , Bull. Lond.
Math. Soc. 53(3) (2021) 792–797.
[32] D. J. Platt and T. S. Trudgian, The error term in the prime number theorem, Math.
Comput. 90(328) (2021) 871–881.
[33] O. Ramaré, Explicit estimates for the summatory function of Λ(n)/n from the one
of Λ(n), Acta Arith. 159(2) (2013) 113–122.
[34] O. Ramaré, Modified truncated Perron formulae, Ann. Math. Blaise Pascal 23(1)
(2016) 109–128.
[35] A. Selberg, On the normal density of primes in small intervals, and the difference
between consecutive primes, Arch. Math. Naturvid. 47 (1943) 87–105.
Int. J. Number Theory 2023.19:1205-1228. Downloaded from www.worldscientific.com

[36] A. Simonič, Estimates for l-functions in the critical strip under GRH with effective
applications, preprint (2022), arXiv:2201.10771.
[37] G. Tenenbaum, Introduction to Analytic and Probabilistic Number Theory, 3rd edn.,
Graduate Studies in Mathematics, Vol. 163 (American Mathematical Society, Provi-
dence, RI, 2015).
[38] D. Wolke, On the explicit formula of Riemann–von Mangoldt, II, J. London Math.
Soc. 2(3) (1983) 406–416.