Preprint
Article

This version is not peer-reviewed.

Robin’s Inequality at Colossally Abundant Numbers and Nicolas’ Inequality at Their Largest Prime Factor

Submitted:

11 November 2026

Posted:

12 November 2026

You are already at the latest version

Abstract
Robin's criterion states that the Riemann Hypothesis holds if and only if \( \sigma(n) < e^{\gamma} \cdot n \cdot \log\log n \) for every \( n > 5040 \), and Nicolas' criterion states that it holds if and only if \( c(q) = \log(q\#/\varphi(q\#)) - \gamma - \log\log\theta(q) > 0 \) for every prime \( q \), where \( q\# \) is the primorial of \( q \) and \( \theta \) is the Chebyshev function. Let \( N \) be a colossally abundant number, let \( \xi(N) \) be the real number associated with \( N \) by Nicolas, and let \( q \) be the largest prime not exceeding \( \xi(N) \). Writing \( \Delta(N) = \log(\sigma(N)/N) - \gamma - \log\log\log N \), we prove without any hypothesis that, whenever \( \log q \geq 31 \), \( 1.353/(\sqrt{q} \log^{2} q) \leq \Delta(N) - c(q) + 2\sqrt{2}/(\sqrt{q} \log q) \leq 3.970/(\sqrt{q} \log^{2} q) \). The range \( \log q \geq 31 \) contains the prime \( q \) of every colossally abundant counterexample, in accordance with the author's result that the largest prime factor of the least counterexample exceeds \( e^{31.018} \). Thus the functions \( \Delta(N) \) and \( c(q) \) differ by \( 2\sqrt{2}/(\sqrt{q} \log q) \) up to an explicit error of order \( 1/(\sqrt{q} \log^{2} q) \). This comparison is two-sided, but it is not an equivalence between Robin's inequality at \( N \) and Nicolas' inequality at \( q \): it does not decide Robin's inequality at \( N \) when \( c(q) \sqrt{q} \log q \) lies between \( 2\sqrt{2}(1 - 1.4035/\log q) \) and \( 2\sqrt{2}(1 - 0.4787/\log q) \). As consequences, every colossally abundant counterexample to Robin's inequality yields a prime \( q \) with \( c(q) > 2.70/(\sqrt{q} \log q) \); conversely, every prime \( q \) with \( \log q \geq 31 \) and \( c(q) > 2\sqrt{2}(1 - 0.4787/\log q)/(\sqrt{q} \log q) \) yields a colossally abundant counterexample, which gives a derivation of Robin's known theorem on colossally abundant counterexamples from Nicolas' oscillation theorem; and under the Riemann Hypothesis every colossally abundant number with \( \log q \geq 31 \) satisfies \( \Delta(N) < -0.654/(\sqrt{q} \log q) \), in accordance with the constant \( 2(\sqrt{2}-1) \) of Ramanujan's bound. For superabundant counterexamples, using the exponent bound of Nazardonyavi and Yakubovich, which gives \( \log n \geq \psi(p_{k}) \) for the largest prime factor \( p_{k} \), we prove \( c(p_{k}) > 0.9991/(\sqrt{p_{k}} \log p_{k}) \), a bound that applies to the least counterexample, which is superabundant, and we show that the least colossally abundant number not smaller than the least counterexample \( n \) is itself a counterexample \( N \) with \( \log N - \log\xi(N) \leq \log n \).
Keywords: 
;  ;  ;  ;  ;  ;  

MSC:  11M26; 11A41; 11A25; 11N37

1. Introduction

The Riemann Hypothesis (RH) is the conjecture that the Riemann zeta function has its zeros only at the negative even integers and at complex numbers with real part 1 2  [1]. It has been called the “Holy Grail of Mathematics” [2,3], and numerous equivalent formulations exist [4]. One of particular interest here is Robin’s criterion [5], which asserts that RH is equivalent to the inequality
σ ( n ) < e γ · n · log log n for every integer n > 5040 .
Here γ ≈ 0.57721 is the Euler–Mascheroni constant, σ ( n ) = ∑ d ∣ n d is the sum-of-divisors function, and log denotes the natural logarithm. Grönwall’s theorem [6] states that lim sup n → ∞ σ ( n ) / ( n log log n ) = e γ , so (1) asserts that this limiting value is never attained beyond 5040. A closely related elementary reformulation in terms of harmonic numbers is due to Lagarias [7], and the extremal integers relevant to both criteria have been studied computationally by Briggs [8] and structurally by Caveney, Nicolas and Sondow [9].
It is known that (1) holds for many classes of integers: odd integers exceeding 9 and squarefree integers exceeding 30 [10], sums of two squares exceeding 720 [11], and t-free integers for t = 5 , 7 , 11 and 20 [10,12,13,14]. Akbary and Friggstad [15] showed that the least counterexample to (1), should one exist, must be a superabundant number, Nazardonyavi and Yakubovich [16] refined this to the smaller class of extremely abundant numbers, and the author proved that its largest prime factor exceeds e 31.018  [17]. Robin [5] proved that if RH is false, then (1) fails for infinitely many colossally abundant numbers.
A second criterion concerns the primorials q # = ∏ p ≤ q p . For a real number x ≥ 2 put
c ( x ) = ∑ p ≤ x log p p − 1 − γ − log log θ ( x ) ,
where θ ( x ) = ∑ p ≤ x log p is the Chebyshev function, so that c ( q ) > 0 is the inequality q # φ ( q # ) > e γ · log log ( q # ) , with φ Euler’s totient function. Nicolas [18] proved that c ( x ) > 0 for every x ≥ 2 if RH is true, and that if RH is false, then c ( x ) = Ω ± ( x − b ) for some 0 < b < 1 2 . Under RH, Ramanujan’s asymptotic upper bound for σ ( n ) / n  [19] contains the term − 2 ( 2 − 1 ) log n , and Nicolas [20] gave an effective form of this bound, using the structure of colossally abundant numbers and his estimate [21] of c ( x ) under RH.
In this paper we compare Robin’s inequality at a colossally abundant number with Nicolas’ inequality at its largest prime factor, without assuming RH. For an integer n ≥ 16 put
Δ ( n ) = log σ ( n ) n − γ − log log log n ,
so that Robin’s inequality holds for n > 5040 if and only if Δ ( n ) < 0 . To each colossally abundant number N Nicolas associates a real number ξ ( N )  [20] (Definition 4.3); every prime factor of N is at most ξ ( N ) (Section 3). Our main result is the following.
Theorem 1.
Let N be a colossally abundant number, and let q be the largest prime with q ≤ ξ ( N ) . If log q ≥ 31 , then
1.353 q · log 2 q ≤ Δ ( N ) − c ( q ) + 2 2 q · log q ≤ 3.970 q · log 2 q .
The hypothesis log q ≥ 31 of Theorem 1 covers every colossally abundant counterexample to (1). In [17] (Theorem 4.2) the author proved that the largest prime factor p k of the least counterexample satisfies log p k > 31.018 , and Lemma 6 gives log p k > 31.0319 for the largest prime factor of every counterexample; since every prime factor of a colossally abundant number N is at most q, the prime q of a colossally abundant counterexample satisfies log q ≥ 31 .
Thus Robin’s function at N and Nicolas’ function at q differ by 2 2 q log q , up to an explicit error of order 1 q log 2 q , that is, smaller by a factor log q . The comparison is two-sided, but it is not an equivalence between Robin’s inequality at N and Nicolas’ inequality at q. Theorem 1 gives Δ ( N ) ≥ 0 when c ( q ) q log q ≥ 2 2 1 − 0.4787 log q and Δ ( N ) < 0 when c ( q ) q log q < 2 2 1 − 1.4035 log q , and it does not determine the sign of Δ ( N ) when c ( q ) q log q lies between these two numbers. Remark 2 lists the terms of the proof that produce this interval.
The relation of Theorem 1 to earlier work is the following. The comparison of σ ( N ) / N with the Mertens product ∏ p ≤ ξ ( N ) p p − 1 at colossally abundant numbers goes back to Robin [5] (Section 4), and the decomposition used here is that of Nicolas’ proof of [20] (Proposition 5.1), which gives, under RH and for ξ ( N ) ≥ 10 9 + 7 , an upper and a lower bound for σ ( N ) / N in terms of log N . In that proof RH is used in two ways. First, the Mertens product is evaluated by the estimate of c ( x ) under RH [21] (2.16), [20] (1.11) (Proposition 8). Second, the numbers log N , θ ( ξ ( N ) ) and ξ ( N ) are compared in [20], which is stated under RH and whose proof uses the bounds | ψ ( x ) − x | ≤ x log 2 x 8 π and | θ ( x ) − x | ≤ x log 2 x 8 π valid under RH [20] (2.7), (2.8); the comparison of log log θ ( ξ ( N ) ) with log log log N in [20] (Lemma 4.17) is deduced from that lemma. The estimates of the sums over the primes p ≤ ξ ( N ) according to their exponent in [20] (5.5)–(5.8) rest on bounds that do not assume RH. In Theorem 1 the Mertens product is not evaluated; it is kept inside c ( q ) . The comparison of log log log N with log log θ ( ξ ( N ) ) is made directly in Lemma 4, by the mean value theorem applied to the difference log N − θ ( ξ ( N ) ) , in which θ ( ξ 2 ( N ) ) is estimated by the bounds of Büthe [22] and Dusart [23], which give | θ ( t ) − t | ≤ 0.022 log ξ ( N ) · t for every t ≥ ξ 2 ( N ) when log ξ ( N ) ≥ 31 (Lemma 1), and θ ( ξ ( N ) ) by the bound of Dusart. The sum ∑ ξ 2 < p ≤ ξ 1 p 2 is estimated by Stieltjes integration (Lemmas 2 and 3), as in [20] (Lemma 2.1), and the primes with exponent at least 2 are treated as in [20] (5.5), (5.6) (Lemma 5). By Proposition 1, a counterexample to (1) exists only if RH is false, so an estimate proved under RH does not apply to a counterexample; Theorem 1 applies to every colossally abundant number with log q ≥ 31 , whether RH holds or not, and it is used in this way in Corollaries 1, 2 and 3. Under RH and for ξ ( N ) ≥ 10 9 + 7 , the estimates of [20] give a narrower interval than Theorem 1; Remark 1 compares the constants.
Theorem 1 has the following consequences.
(1)
Every colossally abundant counterexample to (1) yields a prime q with log q > 31.03 and c ( q ) > 2.70 q log q (Corollary 1).
(2)
Conversely, every prime q with log q ≥ 31 and c ( q ) > 2 2 q log q 1 − 0.4787 log q yields a colossally abundant counterexample to (1) (Corollary 2).
(3)
Nicolas’ theorem c ( x ) = Ω + ( x − b ) under the failure of RH implies, through (2), Robin’s theorem that (1) then fails for infinitely many colossally abundant numbers (Corollary 3).
(4)
Under RH, every colossally abundant number N with log q ≥ 31 satisfies Δ ( N ) < − 0.654 q log q (Corollary 4); the constant 2 2 − 2 = 2 ( 2 − 1 ) obtained from Theorem 1 and the main term of c ( q ) is the constant of Ramanujan’s bound.
(5)
RH holds if and only if c ( q ) < 2.70 q log q for every prime q ≥ 10 9 (Corollary 5). This equivalence also follows from Nicolas’ theorem [18] together with his estimate under RH [20,21], and holds with 2.70 replaced by any constant greater than 2 + τ = 2.0461 … ; Corollary 5 shows that the primes q of item (1) violate it.
For superabundant counterexamples, which need not be colossally abundant, we prove the following (Section 5).
(6)
Every superabundant number n with largest prime factor p k satisfies log n ≥ ψ ( p k ) (Lemma 7, following [16]), and every superabundant counterexample satisfies c ( p k ) > 0.9991 p k log p k (Theorem 2).
(7)
If n is the least counterexample, then the least colossally abundant number N ≥ n is a counterexample, and log N − log ξ ( N ) ≤ log n ≤ log N (Corollary 6).
The status of these statements with respect to the literature is as follows. Items (3) and (5) are derivations of known results: item (3) recovers Robin’s theorem (Proposition 6) from Nicolas’ theorem (Proposition 7) by means of Theorem 1, and the equivalence of item (5) also follows from the results of Nicolas, as stated there. The first claim of item (6) is due to Nazardonyavi and Yakubovich [16], and item (7) is obtained by combining two results of Robin and Nicolas (Proposition 10 (f) and (g)) with item (1). Theorem 1, items (1), (2) and (4), and the second claim of item (6) are proved in this paper.
The proofs use the description of colossally abundant numbers by Alaoglu and Erdos [24], Robin [5] and Nicolas [20], explicit estimates for the Chebyshev function of Rosser and Schoenfeld [25], Büthe [22] and Dusart [23], the estimate of Axler [26] for the Mertens product, and the numerical verification of (1) for all 5040 < n ≤ 10 10 13.11485 due to Morrill and Platt [14]. Section 2 collects these results, Section 3 proves Theorem 1, Section 4 derives its consequences, Section 5 treats superabundant counterexamples, and Section 6 describes the computer verification of the numerical constants.

2. Background

We write I ( n ) = σ ( n ) n for the abundancy index, and we say that Robin’s inequality holds for n if I ( n ) < e γ · log log n . An integer n > 5040 for which Robin’s inequality does not hold is called a counterexample. We write ψ ( x ) = ∑ p j ≤ x log p = ∑ j ≥ 1 θ ( x 1 / j ) for the second Chebyshev function, so that ψ ( x ) = ∑ q ≤ x log x log q log q and ψ ( x ) ≥ θ ( x ) + θ ( x ) . For a prime p we write p # = ∏ q ≤ p q , so that log ( p # ) = θ ( p ) , and p 1 = 2 < p 2 = 3 < … denotes the sequence of primes. Since ∑ p ≤ x log p p − 1 and θ ( x ) change only at primes, c ( x ) = c ( q ) when q is the largest prime with q ≤ x .

2.1. Robin’s Criterion and Counterexamples

Proposition 1.
Robin’s inequality holds for all natural numbers n > 5040 if and only if the Riemann Hypothesis is true [5].
Proposition 2.
For n > 1 , I ( n ) < ∏ q ∣ n q q − 1 = n ′ φ ( n ′ ) , where n ′ is the product of the distinct primes dividing n [10] ((2.7), p. 362).
Proposition 3.
Robin’s inequality holds for every integer n with 5040 < n ≤ 10 10 13.11485  [14] (Theorem 5).
A natural number n is superabundant if I ( m ) < I ( n ) for every m < n , and colossally abundant if there is ε > 0 with σ ( m ) m 1 + ε ≤ σ ( n ) n 1 + ε for every m ≥ 1  [24]; such a number ε is called a parameter of n. Every colossally abundant number is superabundant [24].
Proposition 4.
If n is superabundant with largest prime factor p k , then n = ∏ i = 1 k p i a i with a 1 ≥ a 2 ≥ … ≥ a k ≥ 1  [24] (Theorem 1, p. 450).
Proposition 5.
If n > 5040 is the smallest integer such that Robin’s inequality does not hold for n, then n is superabundant [15].
Proposition 6.
If the Riemann Hypothesis is false, then Robin’s inequality fails for infinitely many colossally abundant numbers [5] (Section 4, Proposition 1).

2.2. Nicolas’ Criterion

Proposition 7.
If the Riemann Hypothesis is true, then c ( x ) > 0 for every x ≥ 2 . If the Riemann Hypothesis is false, then there is b with 0 < b < 1 2 such that c ( x ) = Ω ± ( x − b ) ; more precisely, this holds for every b with 1 − Θ < b < 1 2 , where Θ is the supremum of the real parts of the zeros of the Riemann zeta function [18] (Theorem 3 (b), (c)), see also [21].
In [18,21] the statement is given for log f ( x ) = − c ( x ) , where f ( x ) = e γ · log θ ( x ) · ∏ p ≤ x 1 − 1 p . Put
τ = 2 + γ − log ( 4 π ) = 0.0461914 … , S 1 ( x ) = ∑ ρ x ρ − 1 ρ ( 1 − ρ ) ,
where ρ runs over the nontrivial zeros of the Riemann zeta function [20] (1.3), (1.5).
Proposition 8.
Assume the Riemann Hypothesis. Then | S 1 ( x ) | ≤ τ x for x > 1  [20] (1.4), and for x ≥ 10 9  [21] (2.16), [20] (1.11),
0.055 x log 2 x ≤ − c ( x ) + 2 x log x + S 1 ( x ) log x ≤ 2.062 x log 2 x .
Consequently c ( x ) < 2 + τ x log x < 2.0462 x log x for every x ≥ 10 9 .
Proof. 
The left inequality gives c ( x ) ≤ 2 x log x + S 1 ( x ) log x − 0.055 x log 2 x < 2 + τ x log x , and 2 + τ < 2.0462 .    □

2.3. Explicit Estimates for Primes

Proposition 9.
The following hold.
(a)
θ ( x ) < x for 0 < x ≤ 10 19 , and θ ( x ) > x − 1.94 x for 1427 < x ≤ 10 19  [22] (Theorem 2), see also [20] (2.1), (2.2).
(b)
| θ ( x ) − x | < x log 3 x for x ≥ 89 , 967 , 803  [23] (Theorem 4.2).
(c)
For every x ≥ 89 , 693 there is a prime p with x < p ≤ x · 1 + 1 log 3 x  [23] (Proposition 5.4).
(d)
θ ( x ) ≥ 0.945 · x for x ≥ 853  [25] (Theorem 10), and π ( x ) < 1.26 · x log x for x > 1  [25] (3.6), where π ( x ) is the number of primes p ≤ x .
(e)
∑ p ≥ x 1 p 3 ≤ 0.086 x 2 for x ≥ 1418  [20] (2.15).
(f)
For x ≥ 46 , 909 , 038  [26] (Proposition 6.1),
∏ p ≤ x p p − 1 < e γ · log x · 1 − ε ( log x ) − 1 , ε ( P ) = 1 20 P 3 + 3 16 P 4 .
Lemma 1.
The following hold.
(a)
θ ( t ) < 1.000012 · t for every t > 0 .
(b)
Let λ ≥ 31 . Then | θ ( t ) − t | ≤ 0.022 λ · t for every t ≥ 1.39838 · e λ / 2 .
Proof. (a) For t ≤ 10 19 this is Proposition 9 (a); for t > 10 19 , Proposition 9 (b) gives θ ( t ) < t · 1 + 1 log 3 ( 10 19 ) < 1.000012 · t .
(b) Let t ≥ 1.39838 · e λ / 2 . If t ≤ 10 19 , Proposition 9 (a) gives | θ ( t ) − t | < 1.94 t · t and t ≥ 1.18253 · e λ / 4 , so λ · | θ ( t ) − t | t < 1.94 · λ · e − λ / 4 1.18253 ; the right-hand side is decreasing for λ ≥ 4 , and is less than 0.022 at λ = 31 . If t > 10 19 , Proposition 9 (b) gives λ · | θ ( t ) − t | t < λ log 3 t with log t > max log ( 10 19 ) , λ 2 . If λ ≤ 2 log ( 10 19 ) , this is at most 2 log ( 10 19 ) log 3 ( 10 19 ) < 0.022 , and otherwise it is at most 8 λ 2 < 8 31 2 < 0.022 .    □
Lemma 2.
Let y > 1 and 0 ≤ δ < 1 be such that | θ ( t ) − t | ≤ δ t for every t ≥ y . Put A j = 1 y log j y . Then
( 1 − 3 δ ) · A 1 − A 2 ≤ ∑ p > y 1 p 2 ≤ ( 1 + 3 δ ) · A 1 − ( 1 + δ ) · A 2 + 2 ( 1 + δ ) · A 3 .
Proof. 
Put J j = ∫ y ∞ d t t 2 log j t . Since d d t − 1 t log j t = 1 t 2 log j t + j t 2 log j + 1 t , we have J j = A j − j J j + 1 , hence J 1 = A 1 − J 2 , J 2 = A 2 − 2 J 3 , and 0 < J j ≤ A j . By Stieltjes integration,
∑ p > y 1 p 2 = ∫ y ∞ d θ ( t ) t 2 log t = − θ ( y ) y 2 log y + ∫ y ∞ θ ( t ) · 2 log t + 1 t 3 log 2 t d t ,
and ∫ y ∞ 2 log t + 1 t 2 log 2 t d t = 2 J 1 + J 2 = 2 A 1 − J 2 . Using ( 1 − δ ) t ≤ θ ( t ) ≤ ( 1 + δ ) t for t ≥ y , the sum is at least − ( 1 + δ ) A 1 + ( 1 − δ ) ( 2 A 1 − A 2 ) ≥ ( 1 − 3 δ ) A 1 − A 2 , and at most − ( 1 − δ ) A 1 + ( 1 + δ ) ( 2 A 1 − A 2 + 2 A 3 ) , using J 2 ≥ A 2 − 2 A 3 .    □

3. Colossally Abundant Numbers

3.1. The Structure of Colossally Abundant Numbers

We recall the description given by Nicolas [20] (Section 4), which goes back to Alaoglu and Erdos [24] and Robin [5]. For real t > 1 and an integer j ≥ 1 put
F ( t , j ) = log 1 + 1 t + t 2 + … + t j log t ,
which is decreasing in t. Let N be a colossally abundant number and κ its largest parameter. The number κ determines a sequence of real numbers ξ = ξ 1 , ξ 2 , ξ 3 , … , which are the thresholds at which the exponent of a prime changes (Proposition 10 (a)): ξ = ξ ( N ) > 1 is defined by F ( ξ , 1 ) = κ , and ξ j = ξ j ( N ) for j ≥ 2 by F ( ξ j , j ) = κ , with ξ 1 = ξ ; K is the largest j with ξ j ≥ 2  [20] (Definition 4.3). Put N κ = ∏ j = 1 K ∏ ξ j + 1 < p ≤ ξ j p j . Throughout, the letter ξ without index denotes ξ 1 = ξ ( N ) , and the other terms of the sequence always carry their index j ≥ 2 . The numbers N and N κ have the same largest parameter κ , hence the same numbers ξ , ξ j and K; they need not be equal, and by Proposition 10 (a) and (b) below, N divides N κ and N κ N ≤ ξ . The exponents of the primes are described for N κ , and the results are stated for N.
Proposition 10.
With this notation, the following hold.
(a)
N κ is colossally abundant with largest parameter κ, and N divides N κ ; every prime dividing N is at most ξ, and a prime p ≤ ξ has exponent j in N κ if ξ j + 1 < p ≤ ξ j  [20] ((4.11), Proposition 4.2, Lemma 4.4), [24] (Theorem 10).
(b)
∑ j = 1 K θ ( ξ j ) − log ξ ≤ log N ≤ ∑ j = 1 K θ ( ξ j ) , and ξ ξ + 1 · I ( N κ ) ≤ I ( N ) ≤ I ( N κ ) , where
I ( N κ ) = ∏ j = 1 K ∏ ξ j + 1 < p ≤ ξ j 1 − p − ( j + 1 ) 1 − p − 1
[20] (Proposition 4.5).
(c)
ξ 1 / j ≤ ξ j ≤ ( j ξ ) 1 / j for every j ≥ 2  [20] (4.26).
(d)
2 ξ · 1 − log 2 2 log ξ ≤ ξ 2 for ξ ≥ 1530  [5] (p. 190), [20] (4.28) and ξ 2 ≤ 2 ξ · 1 − 0.323 log ξ for ξ ≥ 10 9  [20] (4.29).
(e)
If ξ > 10 9 + 7 , then ∑ j = 3 K ξ j ≤ 1.9769 · ξ 1 / 3  [20] (Lemmas 4.8–4.10).
(f)
For every integer m ≥ 2 there are two consecutive colossally abundant numbers N ′ < N with N ξ ( N ) ≤ N ′ < m ≤ N  [20] (Proposition 4.6).
(g)
If 3 ≤ N ′ < N are consecutive colossally abundant numbers and N ′ ≤ m ≤ N , then
I ( m ) log log m ≤ max I ( N ′ ) log log N ′ , I ( N ) log log N
[5] (Proposition 1), see also [16].
In (e), Nicolas states [20] (Lemma 4.10) for N ≥ N ( 0 ) , where N ( 0 ) is the colossally abundant number with ξ ( N ( 0 ) ) = 10 9 + 7 . If ξ > 10 9 + 7 , then κ < F ( 10 9 + 7 , 1 ) , and since the parameters of colossally abundant numbers decrease as the numbers increase [20] (4.4), we have N > N ( 0 ) , so (e) applies.

3.2. Three Estimates

Throughout this subsection, N is a colossally abundant number with λ = log ξ ≥ 31 , where ξ = ξ ( N ) , and we put
M ξ = 2 ξ · λ , L ( t ) = log log t , S 2 = ∑ ξ 2 < p ≤ ξ 1 p 2 .
The proof of Theorem 1 rests on three estimates, each with main term M ξ or a multiple of it. Lemma 3 estimates S 2 : by Lemma 2, S 2 has the main term 1 ξ 2 log ξ 2 , and 2 ξ · λ 2 = 1 M ξ . Lemma 4 estimates L ( log N ) − L ( θ ( ξ ) ) : by the mean value theorem this difference has the main term log N − θ ( ξ ) ξ λ , where log N − θ ( ξ ) has the main term θ ( ξ 2 ) , which is 2 ξ up to a factor 1 + O 1 λ , and 2 ξ ξ λ = M ξ . Lemma 5 shows that log I ( N ) is ∑ p ≤ q log p p − 1 − S 2 up to a term of smaller order. The three lemmas give Δ ( N ) − c ( q ) = − 2 M ξ up to terms of the form C · M ξ λ , and the decimal constants in their statements are the values of C obtained for λ ≥ 31 ; these constants are verified in Section 6 (Table 4).
By Proposition 10 (d), and since 1 − log 2 2 · 31 > 0.98882 and 2 · 0.98882 > 1.39838 ,
ξ 2 − : = 2 ξ · 1 − log 2 2 λ ≤ ξ 2 ≤ ξ 2 + : = 2 ξ · 1 − 0.323 λ , ξ 2 ≥ 1.39838 · e λ / 2 ,
so Lemma 1 (b) applies to every t ≥ ξ 2 with δ = 0.022 λ . Table 1 collects the notation of this section. Here ξ 2 − and ξ 2 + are the lower and the upper bound for ξ 2 of Proposition 10 (d); they are used only through (2). For ℓ = log ξ 2 we have ℓ ≤ 1 2 ( λ + log 2 ) by ξ 2 ≤ 2 ξ , and since − log 1 − log 2 2 · 31 < 0.011306 , also ℓ ≥ 1 2 ( λ + log 2 ) − 0.011306 , so that
λ 2 · 1 + 0.670535 λ ≤ ℓ = log ξ 2 ≤ λ 2 · 1 + log 2 λ .
Lemma 3.
M ξ · 1 − 2.43616 λ ≤ S 2 ≤ M ξ · 1 − 1.920374 λ .
Proof. 
Put ℓ = log ξ 2 and A j = 1 ξ 2 ℓ j . Since t ↦ t log t is increasing, the bounds (2) on ξ 2 and (3) on ℓ give
A 1 ≥ M ξ 1 − 0.323 λ 1 + log 2 λ ≥ M ξ 1 + 0.370148 λ ≥ M ξ · 1 − 0.370148 λ
and, since 1 − log 2 2 λ 1 + 0.670535 λ ≥ 1 + 0.316464 λ for λ ≥ 31 ,
A 1 ≤ M ξ 1 + 0.316464 λ ≤ M ξ · 1 − 0.316464 λ + 0 . 316464 2 λ 2 ≤ M ξ · 1 − 0.313233 λ .
Since λ 2 ≤ ℓ ≤ λ 2 1 + log 2 λ , we get A 2 = A 1 ℓ ≤ 2 M ξ λ , A 3 = A 2 ℓ ≤ 4 M ξ λ 2 and
A 2 ≥ 2 M ξ λ · 1 − 0.370148 λ 1 + log 2 λ ≥ 2 M ξ λ · 1 − 1.063296 31 > 1.9314 · M ξ λ .
By Lemma 2 with y = ξ 2 and 3 δ = 0.066 λ ,
∑ p > ξ 2 1 p 2 ≥ 1 − 0.066 λ 1 − 0.370148 λ M ξ − 2 M ξ λ ≥ M ξ · 1 − 2.436148 λ
and
∑ p > ξ 2 1 p 2 ≤ 1 + 0.066 λ 1 − 0.313233 λ M ξ − 1.9314 · M ξ λ + 8.006 · M ξ λ 2 ≤ M ξ · 1 − 1.920374 λ ,
where we used 2 ( 1 + δ ) · 4 < 8.006 and 8.006 λ ≤ 8.006 31 < 0.258259 . By Lemma 2 with y = ξ , using − ( 1 + δ ) A 2 ( ξ ) + 2 ( 1 + δ ) A 3 ( ξ ) ≤ 0 ,
0 ≤ ∑ p > ξ 1 p 2 ≤ 1.0022 ξ λ = 1.0022 2 · ξ · M ξ ≤ 1.0022 · 31 · e − 31 / 2 2 · M ξ λ < 4.08 · 10 − 6 · M ξ λ ,
because λ e − λ / 2 is decreasing for λ ≥ 2 . Since S 2 = ∑ p > ξ 2 1 p 2 − ∑ p > ξ 1 p 2 , the claim follows.    □
Lemma 4.
M ξ · 1 − 0.369659 λ ≤ L ( log N ) − L ( θ ( ξ ) ) ≤ M ξ · 1 − 0.05272 λ .
Proof. 
Put D = log N − θ ( ξ ) . By Proposition 10 (c), ξ 3 ≥ ξ 1 / 3 ≥ e 31 / 3 > 853 , so K ≥ 3 , and by Proposition 9 (d), θ ( ξ 3 ) ≥ 0.945 · ξ 1 / 3 > λ , because 0.945 · e s / 3 > s for s ≥ 31 . Proposition 10 (b) and (e) and Lemma 1 then give
θ ( ξ 2 ) ≤ D ≤ θ ( ξ 2 ) + 1.000012 · 1.9769 · ξ 1 / 3 .
By (2) and Lemma 1 (b),
θ ( ξ 2 ) ≥ 1 − 0.022 λ 1 − log 2 2 λ 2 ξ ≥ 1 − 0.368574 λ 2 ξ
and
θ ( ξ 2 ) ≤ 1 + 0.022 λ 1 − 0.323 λ 2 ξ ≤ 1 − 0.301 λ 2 ξ .
Moreover
1.000012 · 1.9769 · ξ 1 / 3 2 ξ ≤ 1.397905 · e − λ / 6 ≤ 1.397905 · 31 · e − 31 / 6 λ < 0.24717 λ ,
because λ e − λ / 6 is decreasing for λ ≥ 6 . Hence
1 − 0.368574 λ 2 ξ ≤ D ≤ 1 − 0.0538 λ 2 ξ .
By Proposition 9 (b), ξ · ( 1 − λ − 3 ) < θ ( ξ ) < ξ · ( 1 + λ − 3 ) . Hence log N = θ ( ξ ) + D ≤ ξ · 1 + λ − 3 + 2 e − λ / 2 ≤ ξ · 1 + 0.0010488 λ , so that log log N ≤ λ + 0.0010488 λ ≤ λ · 1 + 0.0000339 λ and
log N · log log N ≤ ξ λ · 1 + 0.001085 λ .
Also log θ ( ξ ) ≥ λ + log ( 1 − λ − 3 ) ≥ λ − 1.0001 λ − 3 , so that
θ ( ξ ) · log θ ( ξ ) ≥ ξ λ · ( 1 − λ − 3 ) ( 1 − 1.0001 λ − 4 ) ≥ ξ λ · 1 − 0.00108 λ .
Since L ′ ( t ) = 1 t log t is decreasing for t > 1 , the mean value theorem gives
D log N · log log N ≤ L ( log N ) − L ( θ ( ξ ) ) ≤ D θ ( ξ ) · log θ ( ξ ) .
With 2 ξ ξ λ = M ξ , the lower bound is at least M ξ · 1 − 0.368574 / λ 1 + 0.001085 / λ ≥ M ξ · 1 − 0.369659 λ , and the upper bound is at most M ξ · 1 − 0.0538 / λ 1 − 0.00108 / λ ≤ M ξ · 1 − 0.05272 λ .    □
Lemma 5.
Let q be the largest prime with q ≤ ξ . Then
∑ p ≤ q log p p − 1 − S 2 − 1.0156 · M ξ λ ≤ log I ( N ) ≤ ∑ p ≤ q log p p − 1 − S 2 .
Proof. 
For p ≤ ξ let v p be the exponent of p in N κ . By Proposition 10 (a) and (b), the primes p ≤ ξ are the primes p ≤ q , and
log I ( N κ ) = ∑ p ≤ q log p p − 1 + ∑ p ≤ q log 1 − p − ( v p + 1 ) .
Every term of the second sum is negative, and v p = 1 for ξ 2 < p ≤ ξ ; since log ( 1 − u ) ≤ − u , the second sum is at most − S 2 , and the upper bound follows from I ( N ) ≤ I ( N κ ) .
For the lower bound we split the second sum into three parts. For ξ 2 < p ≤ ξ , − log ( 1 − p − 2 ) ≤ p − 2 1 − ξ 2 − 2 , so this part is at least − S 2 − 2 ξ 2 − 2 S 2 . For ξ 3 < p ≤ ξ 2 , v p = 2 , and by Proposition 9 (e), since ξ 3 > 1418 , this part is at least − 1 1 − ξ 3 − 3 · 0.086 ξ 3 2 ≥ − 0.0861 · ξ − 2 / 3 . For p ≤ ξ 3 , if ξ j + 1 < p ≤ ξ j with j ≥ 3 , then p j + 1 > ξ j + 1 j + 1 ≥ ξ by Proposition 10 (c), so − log ( 1 − p − ( j + 1 ) ) ≤ 1 ξ − 1 ; by Proposition 9 (d) and Proposition 10 (c), the number of such primes is at most π ( 3 ξ ) 1 / 3 < 3 · 1.26 · 3 1 / 3 · ξ 1 / 3 λ < 5.4518 · ξ 1 / 3 λ , so this part is at least − 5.4518 31 · ξ 1 / 3 ξ − 1 ≥ − 0.17587 · ξ − 2 / 3 . Together with I ( N ) ≥ ξ ξ + 1 I ( N κ ) and log 1 + 1 ξ ≤ 1 ξ ≤ e − 31 / 3 · ξ − 2 / 3 ,
log I ( N ) ≥ ∑ p ≤ q log p p − 1 − S 2 − 2 ξ 2 − 2 S 2 − 0.26201 · ξ − 2 / 3 .
Finally ξ − 2 / 3 = λ 2 e − λ / 6 2 · M ξ λ , and λ 2 e − λ / 6 is decreasing for λ ≥ 12 , so 0.26201 · ξ − 2 / 3 ≤ 0.26201 · 31 2 · e − 31 / 6 2 · M ξ λ < 1.0155 · M ξ λ , while 2 ξ 2 − 2 S 2 ≤ 2 ξ 2 − 2 M ξ < 10 − 4 · M ξ λ by Lemma 3 and (2).    □
Table 2 summarizes the proof of Lemma 5: it lists the four terms of log I ( N ) − ∑ p ≤ q log p p − 1 , according to the exponent v p of the prime p in N κ , with the bounds obtained for each of them.

3.3. Proof of Theorem 1

Proof Theorem 1.
Put ξ = ξ ( N ) , λ = log ξ and u = log q ≥ 31 , so that λ ≥ u ≥ 31 . Since q ≤ ξ and the primes p ≤ ξ are the primes p ≤ q , we have θ ( ξ ) = θ ( q ) and
Δ ( N ) − c ( q ) = log I ( N ) − ∑ p ≤ q log p p − 1 − L ( log N ) − L ( θ ( ξ ) ) .
By Lemmas 3, 4 and 5,
− M ξ · 2 − 1.973094 − 1.0156 λ ≤ Δ ( N ) − c ( q ) ≤ − M ξ · 2 − 0.369659 + 2.43616 λ ,
that is, − M ξ · 2 − 0.957494 λ ≤ Δ ( N ) − c ( q ) ≤ − M ξ · 2 − 2.805819 λ . Put M q = 2 q · u . The prime following q exceeds ξ , so by Proposition 9 (c), ξ < q · ( 1 + u − 3 ) and λ < u + u − 3 . Hence M ξ ≤ M q , 1 λ ≥ 1 − u − 4 u , and
M ξ M q ≥ 1 ( 1 + u − 3 ) 1 / 2 ( 1 + u − 4 ) ≥ 1 − u − 2 / 2 + u − 3 u ≥ 1 − 0.000554 u .
Therefore
Δ ( N ) − c ( q ) ≥ − M q · 2 − 0.957494 · ( 1 − 31 − 4 ) u ≥ − M q · 2 − 0.9574 u
and
Δ ( N ) − c ( q ) ≤ − M q · 1 − 0.000554 u 2 − 2.805819 u ≤ − M q · 2 − 2.8070 u .
Since 2 M q = 2 2 q log q , 0.9574 · 2 > 1.353 and 2.8070 · 2 < 3.970 , the theorem follows.    □
In the notation of the proof, Theorem 1 is the pair of inequalities
c ( q ) − 2 2 q log q 1 − 0.4787 log q ≤ Δ ( N ) ≤ c ( q ) − 2 2 q log q 1 − 1.4035 log q .
Remark 1.
The decomposition of Δ ( N ) − c ( q ) used in the proof of Theorem 1 follows Nicolas’ proof of [20] (Proposition 5.1), in which log ( σ ( N ) / N ) is written as U = U 1 + U 2 + U 3 − U 4 − U 5  [20] (5.4), with − U 5 = ∑ p ≤ ξ log p p − 1 the Mertens product; a comparison of σ ( N ) / N with this product for colossally abundant numbers N also underlies Robin’s proof of Proposition 6 [5] (Section 4). In [20] (Proposition 5.1) the Mertens product is evaluated under RH by [21] (2.16), that is, by Proposition 8. In Theorem 1 it is kept in the form c ( q ) + γ + log log θ ( q ) , so that no hypothesis on the zeros of the Riemann zeta function is used, and the remaining terms are bounded from above and from below with explicit constants for every colossally abundant number with log q ≥ 31 . The estimate of the Mertens product that does not assume RH, Proposition 9 (f), is an upper bound with the factor ( 1 − ε ( log x ) ) − 1 , where ε ( log x ) > 1 20 log 3 x , whereas the terms compared in Theorem 1 are of order 1 q log q ; in this paper it is used only in Lemmas 6 and 9. Under RH, [20] contains a comparison of the same form as Theorem 1, with ξ = ξ ( N ) in place of q: for ξ ≥ 10 9 + 7 , the estimates [20] (5.5)–(5.8), (4.68), (4.69) give, for the number W = U − ∑ p ≤ ξ log p p − 1 − log log log N + log log θ ( ξ ) ,
3.0187 ξ log 2 ξ − 0.448 ξ 2 / 3 ≤ W + 2 2 ξ log ξ ≤ 3.844 ξ log 2 ξ + 0.0015 ξ 2 / 3 ,
where U is the sum of [20] (5.4), which satisfies ξ ξ + 1 · U ≤ log σ ( N ) N ≤ U  [20] (5.3), so that W is Δ ( N ) − c ( q ) with log σ ( N ) N replaced by U. Here 3.0187 = 2.5627 + 0.456 and 3.844 = 3.353 + 0.491 are the sums of the constants of [20] (5.8), (4.69) and of [20] (5.7), (4.68), and 0.448 = 0.251 + 0.087 + 0.11 is the sum of the constants of [20] (5.5), (5.6), (4.69); with [20] (1.11) they give the constants 0.9567 = 3.0187 − 2.062 and 3.789 = 3.844 − 0.055 of [20] (5.11), (5.12). The interval from 3.0187 to 3.844 , of length 0.8253 , is contained in the interval from 1.353 to 3.970 of Theorem 1, of length 2.617 . Thus, under RH and for ξ ≥ 10 9 + 7 , the estimates of [20] are more precise than Theorem 1; Theorem 1 holds without RH, and its constants include, at the threshold log q ≥ 31 , the terms of order ξ − 2 / 3 (Remark 2).
Remark 2.
The two bounds of Theorem 1 differ by 3.970 − 1.353 q log 2 q . Consequently, Corollaries 1 and 2 below do not decide whether Robin’s inequality holds at a colossally abundant number N when c ( q ) q log q lies between 2 2 1 − 1.4035 log q and 2 2 1 − 0.4787 log q . In the proof, the two sides differ by ( 2.805819 − 0.957494 ) · M ξ λ = 1.848325 · M ξ λ ; of this, 1.0156 comes from Lemma 5 (the primes p ≤ ξ 2 , which have exponent at least 2), 2.43616 − 1.920374 = 0.515786 from Lemma 3, and 0.369659 − 0.05272 = 0.316939 from Lemma 4. Two of these contributions are bounds, evaluated at the threshold λ = 31 , of quantities that tend to 0 as λ → ∞ . The term 1.0156 · M ξ λ of Lemma 5 bounds a quantity which is at most 0.26201 · λ 2 e − λ / 6 2 + 10 − 4 · M ξ λ (Table 2), and in the proof of Lemma 4 the term 0.24717 λ , which is part of the contribution 0.316939 , bounds 1.397905 · e − λ / 6 , which comes from the primes with exponent at least 3 in log N ; the functions λ 2 e − λ / 6 and λ e − λ / 6 are decreasing for λ ≥ 12 and tend to 0. The rest of the difference comes from the relative error 0.022 λ of θ ( t ) for t ≥ ξ 2 (Lemma 1 (b)) and from the two bounds for ξ 2 of Proposition 10 (d), whose constants are log 2 2 = 0.3465 … and 0.323 , the latter being stated for ξ ≥ 10 9 .

4. Consequences

Lemma 6.
Let n be a counterexample with largest prime factor p k . Then log p k > 31.0319 and log log n < 1.000002 · log p k .
Proof. 
Put L = log log n and P = log p k . By Proposition 3, n > 10 10 13.11485 , so L > 13.11485 · log 10 + log log 10 > 31.032 . By Proposition 2, e γ L ≤ I ( n ) < ∏ p ≤ p k p p − 1 . If p k < 46 , 909 , 038 , then by Proposition 9 (f) at x = 46 , 909 , 038 , this product is less than e γ · 17.6638 · 1.000011 < e γ · 31.032 , a contradiction. Hence p k ≥ 46 , 909 , 038 , and Proposition 9 (f) gives L < P · ( 1 − ε ( P ) ) − 1 . Since ε is decreasing and P ≥ log ( 46 , 909 , 038 ) , ( 1 − ε ( P ) ) − 1 < 1.000011 , so P > 31.032 1.000011 > 31.0316 ; then ( 1 − ε ( P ) ) − 1 ≤ ( 1 − ε ( 31.0316 ) ) − 1 < 1.000002 , so L < 1.000002 · P and P > 31.032 1.000002 > 31.0319 .    □
For the least counterexample, the author proved log p k > 31.018  [17] (Theorem 4.2); Lemma 6 applies to every counterexample.
Corollary 1.
Let N be a colossally abundant counterexample, and q the largest prime with q ≤ ξ ( N ) . Then log q > 31.0319 and
c ( q ) > 2 2 q log q 1 − 1.4035 log q > 2.70 q log q , that is , q # φ ( q # ) > e γ · log θ ( q ) · exp 2.70 q log q .
Proof. 
The largest prime factor p k of N is at most ξ ( N ) by Proposition 10 (a), so p k ≤ q and log q > 31.0319 by Lemma 6. Since Robin’s inequality fails for N, Δ ( N ) ≥ 0 , and the upper bound in (4) gives the first inequality. The function u ↦ 2 2 1 − 1.4035 u is increasing, and its value at u = 31.0319 exceeds 2.70 .    □
Corollary 2.
Let q be a prime with log q ≥ 31 and
c ( q ) > 2 2 q log q 1 − 0.4787 log q .
Then every colossally abundant number N such that q is the largest prime with q ≤ ξ ( N ) is a counterexample. In particular, the colossally abundant number N F ( q , 1 ) of Proposition 10 (a), for which ξ = q , is a counterexample.
Proof. 
The lower bound in (4) gives Δ ( N ) > 0 , that is, I ( N ) > e γ log log N , and N > 5040 because q divides N κ and log N ≥ θ ( q ) − log q by Proposition 10 (b). For κ = F ( q , 1 ) , the number N κ is colossally abundant with largest parameter κ by Proposition 10 (a), and ξ ( N κ ) = q because F ( · , 1 ) is decreasing.    □
By Corollaries 1 and 2, for colossally abundant numbers N with log q ≥ 31 , Robin’s inequality fails at N when c ( q ) q log q > 2 2 1 − 0.4787 log q and holds at N when c ( q ) q log q < 2 2 1 − 1.4035 log q .
Corollary 3.
Proposition 7 and Corollary 2 imply Proposition 6: if the Riemann Hypothesis is false, then Robin’s inequality fails for infinitely many colossally abundant numbers.
Proof. 
If RH is false, Proposition 7 gives b < 1 2 and C > 0 such that c ( x ) > C x − b for arbitrarily large x. Let q be the largest prime with q ≤ x ; then c ( q ) = c ( x ) and q > x 2 , so c ( q ) > C x − b > 4 x log ( x / 2 ) > 2 2 q log q for all sufficiently large such x. Corollary 2 gives a colossally abundant counterexample N F ( q , 1 ) for each such q with log q ≥ 31 , and these numbers are distinct because ξ ( N F ( q , 1 ) ) = q .    □
Remark 3.
Corollary 3 recovers Proposition 6 of Robin [5], here obtained from Proposition 7. Corollary 4 below is an explicit inequality, for colossally abundant numbers with log q ≥ 31 , of the type of Ramanujan’s upper bound for σ ( n ) / n under RH, of which [20] (Theorem 1.1) gives an effective form for every n ≥ 2 .
Corollary 4.
Assume the Riemann Hypothesis. Let N be a colossally abundant number, and q the largest prime with q ≤ ξ ( N ) . If log q ≥ 31 , then
Δ ( N ) < − 0.654 q log q , that is , σ ( N ) N < e γ · log log N · exp − 0.654 q log q .
Proof. 
By (4) and Proposition 8, since q > 10 9 ,
Δ ( N ) q log q < 2 + τ − 2 2 + 2 2 · 1.4035 log q ≤ 2 + τ − 2 2 + 2 2 · 1.4035 31 < − 0.654 .
   □
The number 2 2 − 2 − τ = 2 ( 2 − 1 ) − τ = 0.7822 … appearing in the proof is the constant obtained from Ramanujan’s bound [19] together with | S 1 ( x ) | ≤ τ x , as in [20] (1.8); here q log q plays the role of log N · log log N , since log N = q · ( 1 + o ( 1 ) ) by the proof of Lemma 4 and Proposition 9 (c).
Corollary 5.
The Riemann Hypothesis is true if and only if
q # φ ( q # ) < e γ · log θ ( q ) · exp 2.70 q log q
for every prime q ≥ 10 9 .
Proof. 
The inequality is c ( q ) < 2.70 q log q . If RH is true, it holds for every prime q ≥ 10 9 by Proposition 8. If RH is false, Proposition 6 gives a colossally abundant counterexample, and Corollary 1 gives a prime q with log q > 31.0319 , so q > 10 9 , and c ( q ) > 2.70 q log q .    □
By Propositions 7 and 8, the equivalence of Corollary 5 also holds with 2.70 replaced by any constant greater than 2 + τ . Corollary 1 identifies primes that violate it: the primes q attached to colossally abundant counterexamples.

5. Superabundant Counterexamples

The results of Section 4 concern counterexamples that are colossally abundant, whereas Proposition 5 states that the least counterexample is superabundant. In this section a superabundant counterexample n is treated directly, in terms of its own largest prime factor p k , without assuming that n is colossally abundant. Theorem 2 gives an explicit lower bound for c ( p k ) ; in particular, Nicolas’ inequality c ( p k ) > 0 holds at the largest prime factor of every superabundant counterexample, hence at the largest prime factor of the least counterexample. The constant 0.9991 of Theorem 2 is smaller than the constant 2.70 of Corollary 1, which is available only when the counterexample is colossally abundant, and it is smaller than 2 2 1 − 0.4787 log p k , so Theorem 2 and Corollary 2 do not imply that the colossally abundant number N F ( p k , 1 ) is a counterexample. Corollary 6 relates the least counterexample n to a colossally abundant counterexample N ≥ n , to which Corollary 1 applies with the prime q attached to N. All proofs of this section are given in full, and the numerical constants of Lemmas 8 and 9 and of Theorem 2, in particular 0.9991 , are verified in Section 6 (Table 3).
Lemma 7.
Let n = ∏ i = 1 k p i a i be a superabundant number with largest prime factor p k . Then a i ≥ log p k log p i for every 1 ≤ i ≤ k , and consequently log n ≥ ψ ( p k ) .
Proof. 
The first claim is due to Nazardonyavi and Yakubovich [16]; we give the proof for completeness. For a prime p and an integer b ≥ 0 put g ( p , b ) = ∑ j = 0 b p − j . For i = k the claim is a k ≥ 1 . Let i < k , write q = p i and a = a i ≥ 1 , and suppose that a ≤ log p k log q − 1 . Then q a + 1 ≤ p k , and since q a + 1 ≠ p k , we have q a + 1 < p k . Put m = n · q a + 1 p k , which is an integer with m < n , and write b = a k ≥ 1 . Since I is multiplicative,
I ( n ) I ( m ) = g ( q , a ) g ( q , 2 a + 1 ) · g ( p k , b ) g ( p k , b − 1 ) = 1 + p k − b / g ( p k , b − 1 ) 1 + q − ( a + 1 ) ≤ 1 + 1 p k 1 + 1 q a + 1 < 1 ,
where we used g ( q , 2 a + 1 ) = g ( q , a ) · ( 1 + q − ( a + 1 ) ) , g ( p k , b − 1 ) ≥ 1 and q a + 1 < p k . Thus I ( m ) > I ( n ) with m < n , which contradicts the superabundance of n. Finally, by Proposition 4 the primes dividing n are exactly the primes q ≤ p k , so
log n = ∑ q ≤ p k a q log q ≥ ∑ q ≤ p k log p k log q log q = ψ ( p k ) ,
where a q denotes the exponent of q in n.    □
Lemma 8.
Let n be a counterexample, and n ′ the product of the distinct primes dividing n. Then n ′ > 15 and
log n ′ φ ( n ′ ) − γ − log log log n ′ > log log log n − log log log n ′ ≥ 0 .
Proof. 
By Proposition 2, and since Robin’s inequality fails for n, n ′ φ ( n ′ ) > I ( n ) ≥ e γ · log log n . By Proposition 3, log log n > 31.032 , so n ′ φ ( n ′ ) > 55 . Since m φ ( m ) ≤ 3 for every m ≤ 15 , we get n ′ > 15 , hence log log n ′ > 0 . Taking logarithms gives log n ′ φ ( n ′ ) > γ + log log log n , and subtracting log log log n ′ gives the first inequality. The second follows from n ≥ n ′ .    □
For a superabundant counterexample with largest prime factor p k , Proposition 4 gives n ′ = p k # , so Lemma 8 gives
c ( p k ) > log log log n − log log θ ( p k ) ≥ 0 .
Lemma 9.
Let n be a superabundant counterexample with largest prime factor p k . Then log log n < log p k + 0.0562 log 2 p k and log n · log log n < 1.0000603 · p k · log p k .
Proof. 
Put P = log p k and L = log log n . By Lemma 6, P > 31.0319 , and by Propositions 2 and 9 (f), e γ L ≤ I ( n ) < e γ P · ( 1 − ε ( P ) ) − 1 . Hence
L − P < P · ε ( P ) 1 − ε ( P ) = 1 P 2 · 1 20 + 3 16 P 1 − ε ( P ) ≤ 1 P 2 · 1 20 + 3 16 · 31.0319 1 − ε ( 31.0319 ) < 0.0562 P 2 .
Therefore log n < p k · e 0.0562 / P 2 and L < P · 1 + 0.0562 P 3 , so that
log n · log log n < p k · P · e 0.0562 31 . 03 2 · 1 + 0.0562 31 . 03 3 < 1.0000603 · p k · P .
   □
Theorem 2.
Let n be a superabundant counterexample with largest prime factor p k . Then
c ( p k ) > log log ψ ( p k ) − log log θ ( p k ) > 0.9991 p k · log p k .
Proof. 
By (5) and Lemma 7, c ( p k ) > log log log n − log log θ ( p k ) ≥ log log ψ ( p k ) − log log θ ( p k ) . The function t ↦ log log t has the decreasing derivative 1 t log t for t > 1 , and θ ( p k ) < θ ( p k ) + θ ( p k ) ≤ ψ ( p k ) ≤ log n . By the mean value theorem and Lemma 9,
log log ψ ( p k ) − log log θ ( p k ) ≥ ψ ( p k ) − θ ( p k ) ψ ( p k ) · log ψ ( p k ) ≥ θ ( p k ) log n · log log n > θ ( p k ) 1.0000603 · p k · log p k .
By Lemma 6, p k > e 15.515 > 5.47 · 10 6 , so Proposition 9 (a) gives θ ( p k ) > 1 − 1.94 5.47 · 10 6 p k > 0.99917 · p k if p k ≤ 10 19 , and Proposition 9 (b) gives θ ( p k ) > ( 1 − 0.000012 ) p k otherwise. Since 0.99917 1.0000603 > 0.9991 , the second inequality follows.    □
Remark 4.
The second inequality of Theorem 2 compares two functions of the prime p k alone, and its constant is bounded by 1 in the following sense: for every η > 0 and every sufficiently large prime p,
log log ψ ( p ) − log log θ ( p ) < 1 + η p · log p .
Indeed, ψ ( x ) − θ ( x ) = θ ( x ) + ∑ j ≥ 3 θ ( x 1 / j ) , where the terms with j > log x log 2 vanish, so by Lemma 1 (a) the sum over j ≥ 3 is at most 1.000012 · x 1 / 3 log x log 2 , and Proposition 9 (b) gives θ ( x ) = x · ( 1 + o ( 1 ) ) and θ ( x ) = x · ( 1 + o ( 1 ) ) as x → ∞ ; hence ψ ( x ) − θ ( x ) = x · ( 1 + o ( 1 ) ) . Since 1 t log t is decreasing, the mean value theorem gives log log ψ ( x ) − log log θ ( x ) ≤ ψ ( x ) − θ ( x ) θ ( x ) · log θ ( x ) = 1 + o ( 1 ) x log x .
Corollary 6.
Suppose that the Riemann Hypothesis is false, and let n be the least counterexample. Let N ′ < N be the consecutive colossally abundant numbers with N ′ < n ≤ N . Then N is a counterexample,
log N − log ξ ( N ) ≤ log N ′ < log n ≤ log N ,
and the largest prime q ≤ ξ ( N ) satisfies c ( q ) > 2.70 q log q .
Proof. 
By Proposition 10 (f) applied to m = n , the consecutive colossally abundant numbers N ′ < n ≤ N satisfy N ξ ( N ) ≤ N ′ , which gives the displayed inequalities. The number 55440 is colossally abundant [20] (Figure 1), and 55440 < n by Proposition 3, so N ′ ≥ 55440 > 5040 . Since N ′ < n , Robin’s inequality holds for N ′ , that is, I ( N ′ ) log log N ′ < e γ , whereas I ( n ) log log n ≥ e γ . By Proposition 10 (g), I ( N ) log log N ≥ I ( n ) log log n ≥ e γ , and N ≥ n > 5040 , so N is a counterexample. The last claim is Corollary 1.    □

6. Experimental Verification

Every numerical inequality between explicit constants used in this paper was checked by a self-contained Python 3.12 script, verify_robin_nicolas.py, written with the help of Claude Opus 5.5 (Anthropic) and run by the author. The script uses the arbitrary-precision library mpmath with 60 significant decimal digits, and exact rational arithmetic (Python’s fractions module) for the inequalities that involve only decimal constants, some of which hold with equality. For each inequality it prints the computed value of its left-hand side. The script has three parts, and all 69 checks performed by the script passed: 14 in Part 1, 50 in Part 2 and 5 in Part 3. We now describe the output.

6.1. Part 1: Counterexamples

Part 1 checks the constants of Lemmas 6, 8 and 9 and of Theorem 2. The computed values are listed in Table 3. The starting point is log log 10 10 13.11485 = 31.0320905 … , the bound for log log n coming from Proposition 3. At x = 46 , 909 , 038 , where Proposition 9 (f) begins to apply, log x = 17.6637 … and ( 1 − ε ( log x ) ) − 1 = 1.0000109986 … , which lies below 1.000011 by 1.4 · 10 − 9 ; this is the smallest margin in Part 1. The two steps of Lemma 6 give 31.032 1.000011 = 31.03165 … and 31.032 1.000002 = 31.031937 … . In Lemma 9, the constant 0.0562 exceeds the computed value 0.0560422 … , and the factor 1.0000603 exceeds the computed value 1.0000602505 … by 4.9 · 10 − 8 .

6.2. Part 2: The Proof of Theorem 1

Part 2 checks the constants of Lemmas 1, 3, 4 and 5 and of the proof of Theorem 1, all at the threshold λ = 31 , where e 31 = 29048849665247.4 … ; each quantity checked at λ = 31 is monotone in λ in the direction stated in the proofs. Table 4 lists the main values. In Lemma 1 (b), the relative error 0.022 λ for θ ( t ) on t ≥ ξ 2 comes from Büthe’s bound, with computed value 0.0219063 … at λ = 31 , whereas Dusart’s bound contributes at most 0.00105 . In Lemma 4, the contribution 1.397905 · 31 · e − 31 / 6 = 0.2471636 … of the primes with exponent at least 3 reduces the coefficient 0.301 coming from ξ 2 to 0.0538 . In Lemma 5, the primes with exponent 2 and those with exponent at least 3 contribute 0.262002530 … · ξ − 2 / 3 , which is 1.0154801 … · M ξ λ at λ = 31 . Twelve of the checks in Part 2 are identities or inequalities between the decimal constants of the proofs, such as 0.066 + 0.370148 + 2 = 2.436148 and 0.313233 − 0.066 + 1.9314 − 0.258259 = 1.920374 ; they are verified in exact rational arithmetic. The smallest positive margins in Part 2 are those of 1 2 · 31 2 + 1 31 3 = 0.000553858 … < 0.000554 , of 2 log 2 2 − 0.011306 = 0.6705351805 … ≥ 0.670535 , and of 2 1 − 0.370148 + log 2 31 = 1.9314003 … > 1.9314 , with differences 1.4 · 10 − 7 , 1.8 · 10 − 7 and 3.1 · 10 − 7 . The final constants of Theorem 1 are 0.9574 · 2 = 1.3539680 … > 1.353 and 2.8070 · 2 = 3.9696974 … < 3.970 .
Table 4. Values computed in Part 2 of the script, at λ = 31 .
Table 4. Values computed in Part 2 of the script, at λ = 31 .
Quantity Computed value Bound used
1 − log 2 62 0.9888202067 … > 0.98882
1.94 · 31 · e − 31 / 4 / 1.18253 0.0219062995 … < 0.022
− log 1 − log 2 62 0.0112427568 … < 0.011306
0.670535 − log 2 2 − log 2 2 · 0.670535 31 0.3164649670 … > 0.316464
0.316464 − 0 . 316464 2 / 31 0.3132333721 … > 0.313233
2 1 − 0.370148 + log 2 31 1.9314003109 … > 1.9314
1.0022 · 31 · e − 31 / 2 / 2 4.0760229901 … · 10 − 6 < 4.08 · 10 − 6
1.397905 · 31 · e − 31 / 6 0.2471636095 … < 0.24717
1 31 2 + 2 · 31 · e − 31 / 2 0.0010487169 … < 0.0010488
0.0861 + 0.17587 + e − 31 / 3 0.2620025305 … < 0.26201
0.26201 · 31 2 · e − 31 / 6 / 2 1.0154801390 … < 1.0155
1 2 · 31 2 + 1 31 3 0.0005538585 … < 0.000554
0.957494 · ( 1 − 31 − 4 ) 0.9574929632 … > 0.9574
0.9574 · 2 1.3539680646 … > 1.353
2.8070 · 2 3.9696974696 … < 3.970

6.3. Part 3: The Corollaries

Part 3 checks the constants of Proposition 8 and Corollaries 1, 4 and 5. The script computes τ = 2 + γ − log ( 4 π ) = 0.0461914179 … , so that 2 + τ = 2.0461914179 … < 2.0462 , and 2 2 − 2 − τ = 0.7822357068 … , the constant discussed after Corollary 4. The constant of Corollary 1 is 2 2 1 − 1.4035 31.0319 = 2.7005040047 … > 2.70 , and the constant of Corollary 4 is 2 + τ − 2 2 + 2 2 · 1.4035 31 = − 0.6541809497 … < − 0.654 . The constant 2.70 of Corollary 5 exceeds the bound 2.0462 of Proposition 8 by 0.6538 . Finally, e 31.0319 = 2.999 … · 10 13 > 10 9 , so the primes q of Corollary 1 lie in the range of Proposition 8.

Supplementary Materials

The following supporting information can be downloaded at the website of this paper posted on Preprints.org.

Data Availability Statement

The Python script verify_robin_nicolas.py used in Section 6 is available in the preprint.

Acknowledgments

This work was carried out with the help of Claude Opus 5.5 (Anthropic): the verification script verify_robin_nicolas.py used in Section 6 and parts of the text of this paper were generated with this AI assistant. The experiments were run by the author.

Conflicts of Interest

The author declares no conflicts of interest.

References

  1. Borwein, P.B.; Choi, S.; Rooney, B.; Weirathmueller, A. The Riemann Hypothesis: A Resource for the Afficionado and Virtuoso Alike . In CMS Books in Mathematics; Springer: New York, 2008; Vol. 27. [Google Scholar]
  2. Conrey, B. Conrey, B., Riemann’s hypothesis. In <italic>Colloquium De Giorgi 2013 and 2014</italic>; Publications of the Scuola Normale Superiore ((COLLOQUIASNS,volume 5)), Scuola Normale Superiore, 2015; pp. 109–117. [CrossRef]
  3. Connes, A. An Essay on the Riemann Hypothesis. Open Probl. Math. 2016, 225–257. [Google Scholar] [CrossRef]
  4. Conrey, B. The Riemann Hypothesis. Not. AMS 2003, 50, 341–353. [Google Scholar]
  5. Robin, G. Grandes valeurs de la fonction somme des diviseurs et hypothèse de Riemann. J. De Mathématiques Pures Et. Appliquées. Neuvième Série 1984, 63, 187–213. [Google Scholar]
  6. Grönwall, T.H. Some asymptotic expressions in the theory of numbers. Trans. Am. Math. Soc. 1913, 14, 113–122. [Google Scholar] [CrossRef]
  7. Lagarias, J.C. An elementary problem equivalent to the Riemann hypothesis. Am. Math. Mon. 2002, 109, 534–543. [Google Scholar] [CrossRef]
  8. Briggs, K. Abundant numbers and the Riemann hypothesis. Exp. Math. 2006, 15, 251–256. [Google Scholar] [CrossRef]
  9. Caveney, G.; Nicolas, J.L.; Sondow, J. On SA, CA, and GA numbers. Ramanujan J. 2012, 29, 359–384. [Google Scholar] [CrossRef]
  10. Choie, Y.; Lichiardopol, N.; Moree, P.; Solé, P. On Robin’s criterion for the Riemann hypothesis. J. De Théorie Des. Nr. De Bordx. 2007, 19, 357–372. [Google Scholar] [CrossRef]
  11. Banks, W.D.; Hart, D.; Moree, P.; Nevans, C.W. The Nicolas and Robin inequalities with sums of two squares. Monatshefte Für Math. 2009, 157, 303–322. [Google Scholar] [CrossRef]
  12. Solé, P.; Planat, M. The Robin inequality for 7-free integers. Integers 2012, 12, 301–309. [Google Scholar] [CrossRef]
  13. Broughan, K.; Trudgian, T. Robin’s inequality for 11-free integers. Integers 2015, 15. [Google Scholar]
  14. Platt, D.J.; Morrill, T. Robin’s inequality for 20-free integers. Integers 2021, 21. [Google Scholar]
  15. Akbary, A.; Friggstad, Z. Superabundant numbers and the Riemann hypothesis. Am. Math. Mon. 2009, 116, 273–275. [Google Scholar] [CrossRef]
  16. Nazardonyavi, S.; Yakubovich, S. Extremely abundant numbers and the Riemann hypothesis. J. Integer Seq. 2014, 17. [Google Scholar]
  17. Vega, F. Robin’s criterion on divisibility. Ramanujan J. 2022, 59, 745–755. [Google Scholar] [CrossRef]
  18. Nicolas, J.L. Petites valeurs de la fonction d’Euler. J. Number Theory 1983, 17, 375–388. [Google Scholar] [CrossRef]
  19. Ramanujan, S. Ramanujan, S. Highly composite numbers. The Ramanujan Journal 1997, 1, 119–153. Annotated and with a foreword by Jean-Louis Nicolas and Guy Robin. [CrossRef]
  20. Nicolas, J.L. The sum of divisors function and the Riemann hypothesis. Ramanujan J. 2022, 58, 1113–1157. [Google Scholar] [CrossRef]
  21. Nicolas, J.L. Small values of the Euler function and the Riemann hypothesis. Acta Arith. 2012, 155, 311–321. [Google Scholar] [CrossRef]
  22. Büthe, J. An analytic method for bounding ψ(x). Math. Comput. 2018, 87, 1991–2009. [Google Scholar] [CrossRef]
  23. Dusart, P. Explicit estimates of some functions over primes. Ramanujan J. 2018, 45, 227–251. [Google Scholar] [CrossRef]
  24. Alaoglu, L.; Erdos, P. On highly composite and similar numbers. Trans. Am. Math. Soc. 1944, 56, 448–469. [Google Scholar] [CrossRef]
  25. Rosser, J.B.; Schoenfeld, L. Approximate formulas for some functions of prime numbers. Ill. J. Math. 1962, 6, 64–94. [Google Scholar] [CrossRef]
  26. Axler, C. New estimates for some functions defined over primes. Integers 2018, 18. [Google Scholar]
Table 1. Notation of Section 3.
Table 1. Notation of Section 3.
Symbol Meaning
κ Largest parameter of the colossally abundant number N.
ξ = ξ 1 , ξ j Real numbers with F ( ξ j , j ) = κ ; a prime p with ξ j + 1 < p ≤ ξ j
has exponent j in N κ .
K Largest j with ξ j ≥ 2 .
N κ Colossally abundant number ∏ j = 1 K ∏ ξ j + 1 < p ≤ ξ j p j ,
a multiple of N.
q, u Largest prime with q ≤ ξ , and u = log q .
λ log ξ .
M ξ , M q 2 ξ · λ and 2 q · u .
L ( t ) log log t .
S 2 ∑ ξ 2 < p ≤ ξ 1 p 2 .
D log N − θ ( ξ ) .
A j 1 y log j y (Lemma 2),
with y = ξ 2 and y = ξ in the proof of Lemma 3.
Table 2. The terms of log I ( N ) − ∑ p ≤ q log p p − 1 in the proof of Lemma 5.
Table 2. The terms of log I ( N ) − ∑ p ≤ q log p p − 1 in the proof of Lemma 5.
Primes Term Upper bound Lower bound
ξ 2 < p ≤ ξ ( v p = 1 ) ∑ log 1 − p − 2 − S 2 − S 2 − 2 ξ 2 − 2 S 2
ξ 3 < p ≤ ξ 2 ( v p = 2 ) ∑ log 1 − p − 3 0 − 0.0861 · ξ − 2 / 3
p ≤ ξ 3 ( v p ≥ 3 ) ∑ log 1 − p − ( v p + 1 ) 0 − 0.17587 · ξ − 2 / 3
all p ≤ ξ log I ( N ) − log I ( N κ ) 0 − e − 31 / 3 · ξ − 2 / 3
Total − S 2 − S 2 − 1.0156 · M ξ λ
Table 3. Values computed in Part 1 of the script.
Table 3. Values computed in Part 1 of the script.
Quantity Computed value Bound used
log log 10 10 13.11485 31.0320905521 … > 31.032
log ( 46 , 909 , 038 ) 17.6637209227 … < 17.6638
( 1 − ε ( log 46 , 909 , 038 ) ) − 1 1.0000109986 … < 1.000011
31.032 / 1.000011 31.0316586518 … > 31.0316
( 1 − ε ( 31.0316 ) ) − 1 1.0000018754 … < 1.000002
31.032 / 1.000002 31.0319379361 … > 31.0319
1 20 + 3 16 · 31.0319 / ( 1 − ε ( 31.0319 ) ) 0.0560422746 … < 0.0562
e 0.0562 / 31 . 03 2 1 + 0.0562 31 . 03 3 1.0000602505 … < 1.0000603
e 15.515 5471153.3376 … > 5.47 · 10 6
1 − 1.94 / 5.47 · 10 6 0.9991705159 … > 0.99917
0.99917 / 1.0000603 0.9991097537 … > 0.9991
e γ · 31.032 55.2702392751 … > 55
max m ≤ 15 m / φ ( m ) 3 = 3
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.