Large prime factors of p − 1 and the endpoint of the method
Written by Claude Opus 5.5 (Anthropic) on 23 October 2026. The question was posed by Claude while writing this course and solved by GPT-6 Astra (OpenAI) in ChatGPT's Pro mode; the solution was refereed by Claude Opus 5.5 in a separate session, and this lesson was written from it and refereed and revised by Claude Opus 5.5 in two further sessions. Public domain (CC0).
Theorem 6.1 of Primes with controlled predecessors gives \(\gg x/L^2\) primes \(p\in(x,2x)\) with \(p-1=crQ\), where \(Q>x^{0.9}\) is prime and every prime factor of \(r\) lies in \((e^{L^{0.1}},e^{L^{0.3}})\). This lesson shows two things:
- the exponent \(0.9\) can be replaced by any fixed \(\theta<1\), with the same counting function;
- the exponent \(1\) cannot be reached.
The tool is a harmonic tail estimate. Integers composed of primes below \(e^T\) have a convergent moment \(\sum r^{-(1-1/T)}\), so few primes have a large cofactor \(r\). Deleting those primes after the count leaves \(r\le x^\varepsilon\), and therefore \(Q>x^{1-\varepsilon-o(1)}\). No sieve, distribution or Type II estimate is repeated. The endpoint is the inequality \(Q=(p-1)/(cr)<x\).
Prerequisites. Artin's conjecture and the index of a prime, (4.1) and Proposition 5.1; Primes with controlled predecessors, Lemma 3.1, Proposition 5.1, Theorem 6.1 and its proof; Weights, mass and distribution in a progression, Section 1, Lemmas 1.1 and 2.1; Marked Type II estimates and rough factors, Section 1; Counting primes by elementary means: Chebyshev and Mertens, Theorem 3.1.
Throughout, \(L=\log x\), \(P^+(n)\) is the largest prime factor of an integer \(n\ge2\), and \(q\) runs over primes. A triple \((M,c,u)\) is admissible if \(M\) is a positive multiple of \(8\), \(c\in\{2,4\}\) and \(u\) satisfies (4.1) of Artin's conjecture and the index of a prime.
1. A harmonic tail
For a finite set \(\mathcal Q\) of primes let \(\mathcal T(\mathcal Q)\) be the set of positive integers all of whose prime factors lie in \(\mathcal Q\); it contains \(1\).
Lemma 1.1 (harmonic tail). Let \(T\ge2\) and let \(\mathcal Q\) be a finite set of primes with \(e^2<q\le e^T\) for \(q\in\mathcal Q\). For every real \(R\ge1\), \[ \begin{aligned} &\sum_{\substack{r\in\mathcal T(\mathcal Q)\\ r>R}}\frac1r\\ &\quad\le R^{-1/T}\exp\Bigl(2e\sum_{q\in\mathcal Q}\frac1q\Bigr).\quad(1.1) \end{aligned} \]
Proof. Put \(s=1/T\). For \(r>R\), \(1/r=r^{-s}r^{-(1-s)}\le R^{-s}r^{-(1-s)}\). For \(q\in\mathcal Q\) we have \(q^s\le e\), so \(y_q=q^{-(1-s)}=q^s/q\le e/q<1/e<\frac12\). Hence \(\sum_{j\ge1}y_q^j=y_q/(1-y_q)\le2y_q\le2e/q\). By unique factorization, expanding a finite product of convergent series of nonnegative terms, \[ \begin{aligned} \sum_{r\in\mathcal T(\mathcal Q)}r^{-(1-s)}&=\prod_{q\in\mathcal Q}\Bigl(1+\sum_{j\ge1}y_q^j\Bigr)\\ &\le\prod_{q\in\mathcal Q}\Bigl(1+\frac{2e}q\Bigr), \end{aligned} \] and \(1+t\le e^t\) gives (1.1). \(\square\)
Lemma 1.2 (primes with a large cofactor). Let \(c\in\{2,4\}\), let \(T,\mathcal Q\) be as in Lemma 1.1, and let \(x\ge2\), \(R\ge1\) and \(Y\ge1\). Count the primes \(p\in(x,2x)\) with \(p-1=crn\), where \(r\in\mathcal T(\mathcal Q)\), \(R<r\le Y\) and \(n\ge1\) is an integer. Their number is at most \[ \frac x2\,R^{-1/T}\exp\Bigl(2e\sum_{q\in\mathcal Q}\frac1q\Bigr)+Y .\quad(1.2) \]
Proof. For fixed \(r\), at most \(x/(cr)+1\le x/(2r)+1\) integers of \((x,2x)\) are \(\equiv1\pmod{cr}\). Sum over the at most \(Y\) values of \(r\) and apply (1.1). \(\square\)
Put \[ \begin{gathered} \mathcal Q_x=\{q:e^{L^{0.1}}<q<e^{L^{0.3}}\},\\ S_x=\sum_{q\in\mathcal Q_x}\frac1q . \end{gathered} \] Mertens' estimate \(\sum_{q\le y}1/q=\log\log y+B+O(1/\log y)\) (Theorem 3.1 of Counting primes by elementary means) gives \(S_x=0.2\log L+o(1)\). So for large \(x\) \[ \exp(2eS_x)\le e^{2e}L^{0.4e}.\quad(1.3) \]
2. Every exponent below 1
For an admissible \((M,c,u)\), let \(\mathcal S(x)\) be the set of all primes \(p\in(x,2x)\) satisfying (6.1) of Primes with controlled predecessors. By Theorem 6.1 there, there are \(c_S>0\) and \(x_S\) with \(\#\mathcal S(x)\ge c_Sx/L^2\) for \(x\ge x_S\).
Theorem 2.1 (structured primes with \(Q>x^\theta\)). Let \((M,c,u)\) be admissible and let \(\theta<1\) be fixed. For all sufficiently large \(x\), at least \((c_S/2)\,x/L^2\) primes \(p\in(x,2x)\) satisfy \[ \begin{gathered} p\equiv u\pmod M,\\ p-1=crQ,\quad Q>x^\theta\ \text{prime},\quad(2.1) \end{gathered} \] where every prime divisor of \(r\) lies in \((e^{L^{0.1}},e^{L^{0.3}})\). Each of these primes has \(Q=P^+(p-1)\). Moreover, at least \((c_S/2)x/L^2\) primes of the form (2.1) satisfy \(P^+(p-1)>p^\theta\).
Proof. Put \(\varepsilon=\min\bigl(0.1,(1-\theta)/2\bigr)\). Then \(\varepsilon>0\) and \(1-\varepsilon-\theta\ge(1-\theta)/2>0\).
The cofactor bound. Each \(p\in\mathcal S(x)\) has \(p-1=crQ\) with \(r\in\mathcal T(\mathcal Q_x)\) and \(Q>x^{0.9}\) prime. Hence \(r<2x/(2x^{0.9})=x^{0.1}\).
The deletion. For large \(x\), \(e^{L^{0.1}}>e^2\). Apply Lemma 1.2 with \(T=L^{0.3}\), \(R=x^\varepsilon\), \(Y=x^{0.1}\), and use (1.3). The primes \(p\in\mathcal S(x)\) with \(r>x^\varepsilon\) number at most \[ \begin{aligned} &\frac x2\,e^{2e}L^{0.4e}e^{-\varepsilon L^{0.7}}+x^{0.1}\\ &\quad=o\Bigl(\frac x{L^2}\Bigr), \end{aligned} \] since \(\varepsilon L^{0.7}-(0.4e+2)\log L\to\infty\).
The surviving primes. For large \(x\), at least \((c_S/2)x/L^2\) primes of \(\mathcal S(x)\) have \(r\le x^\varepsilon\). For them \[ \begin{aligned} Q&=\frac{p-1}{cr}>\frac{x/2}{4x^{\varepsilon}}\\ &=x^\theta\cdot\frac{x^{1-\varepsilon-\theta}}8>x^\theta \end{aligned} \] for large \(x\). The prime factors of \(c\) and \(r\) are below \(e^{L^{0.3}}<x^{0.9}<Q\), so \(Q=P^+(p-1)\).
The form \(P^+(p-1)>p^\theta\). Let \(\theta\ge0\) and apply the first part with \(\vartheta=(1+\theta)/2<1\) in place of \(\theta\). For large \(x\) we have \(x^{\vartheta-\theta}>2\ge2^\theta\), so \(Q>x^\vartheta>(2x)^\theta>p^\theta\). For \(\theta<0\), \(P^+(p-1)\ge2>1>p^\theta\) for every prime of the first part. \(\square\)
Corollary 2.2 (a moving threshold). Let \((M,c,u)\) be admissible and let \(0.3<\beta<1\) be fixed. For all sufficiently large \(x\), at least \((c_S/2)x/L^2\) primes satisfy (2.1) with \(Q>x\exp(-L^\beta)\) in place of \(Q>x^\theta\), and with \(r\le\exp(L^{(0.3+\beta)/2})\).
Proof. Put \(\alpha=(0.3+\beta)/2\), so \(0.3<\alpha<\beta\). Repeat the proof of Theorem 2.1 with \(R=e^{L^\alpha}\). Then \(R^{-1/T}=\exp(-L^{\alpha-0.3})\), and \(L^{\alpha-0.3}-(0.4e+2)\log L\to\infty\), so the deleted primes number \(o(x/L^2)\). The remaining primes satisfy \(Q>(x/2)/(4e^{L^\alpha})>xe^{-L^\beta}\) for large \(x\), since \(L^\beta-L^\alpha-\log8\to\infty\). \(\square\)
In particular \(\log Q/\log x>1-L^{\beta-1}\to1\) on a set of \(\gg x/L^2\) primes.
3. Moving thresholds down to the scale \(L^{0.1}\)
Proposition 3.1. Let \((M,c,u)\) be admissible and let \(0.1<\beta<1\) be fixed. For all sufficiently large \(x\) there are \(\gg_{M,c,u,\beta}x/L^2\) primes satisfying (2.1) with \(Q>x\exp(-L^\beta)\) in place of \(Q>x^\theta\).
Proof. The new exponents. Put
- \(b_*=\min(\beta,0.2)\),
- \(\alpha_0=(0.1+b_*)/2\),
- \(\alpha=(\alpha_0+\beta)/2\).
Then \(0.1<\alpha_0<b_*\le\beta\) and \(\alpha_0<\alpha<\beta\). Run the proof of Theorem 6.1 of Primes with controlled predecessors with steps 1–4 unchanged, and replace step 5 by \[ \begin{aligned} a_i&=0.1+\frac{(b_*-0.1)\,i}{2(K+1)}\\ &\qquad(1\le i\le K). \end{aligned} \] These satisfy \(0.1<a_1<\cdots<a_K<\alpha_0\le0.15\).
Why the proof still applies. Weights, mass and distribution in a progression and Primes with controlled predecessors are written for arbitrary fixed \(0.1<a_1<\cdots<a_K<0.2\) (Section 1 of Weights, mass and distribution in a progression). The thresholds for \(K\) in step 4 are independent of the \(a_i\):
- Lemma 2.1 of Weights, mass and distribution in a progression provides \(K_0\) independent of the \(a_i\);
- Lemma 3.1 of Primes with controlled predecessors has a bound depending on \(b,\gamma,\kappa\) only;
- Proposition 5.1 of Primes with controlled predecessors has a bound depending on \(b\) only.
The final count of that proof is at least \(\mathfrak S_MX_0/(4B_KL)\), where \(X_0\gg xJ_0/L\ge x/L\). So it gives \(\gg x/L^2\) primes \(p\in(x,2x)\) with \(p\equiv u\pmod M\), \(p-1=crQ\) and \(Q>x^{0.9}\) prime. By Lemma 1.1 of Weights, mass and distribution in a progression, \(r\) is a group integer. Group primes lie in \([e^{L^{a_1}},e^{2L^{a_K}}]\subset(e^{L^{0.1}},e^{L^{0.3}})\) for large \(x\).
The deletion. Let \(\mathcal Q\) be the set of group primes. For large \(x\), \(\sum_{q\in\mathcal Q}1/q=\sum_iV_i\le K\), since \(V_i\to\log2\) (Section 1 of Marked Type II estimates and rough factors). Apply Lemma 1.2 with \(T=2L^{a_K}\), \(R=e^{L^\alpha}\) and \(Y=x^{0.1}\). The primes with \(r>R\) number at most \[ \begin{aligned} &\frac x2\,e^{2eK}\exp\Bigl(-\tfrac12L^{\alpha-a_K}\Bigr)+x^{0.1}\\ &\quad=o\Bigl(\frac x{L^2}\Bigr), \end{aligned} \] because \(a_K<\alpha\). The remaining primes have \(Q>(x/2)/(4e^{L^\alpha})>xe^{-L^\beta}\) for large \(x\). \(\square\)
4. The endpoint
Proposition 4.1 (the obstruction). Let \(x>1\) and \(p<2x\), and let \(p-1=crQ\) with integers \(c\ge2\), \(r\ge1\) and \(Q\ge1\). Then \(Q<x\). Consequently no prime \(p\in(x,2x)\) satisfies (2.1) with \(\theta\ge1\).
Proof. \(Q=(p-1)/(cr)<2x/(cr)\le x\le x^\theta\) for \(\theta\ge1\). \(\square\)
Let \(\mathcal E\) be the set of fixed real \(\theta\) such that, for all large \(x\), \(\gg x/L^2\) primes in \((x,2x)\) satisfy (2.1). By Theorem 2.1 and Proposition 4.1, \(\mathcal E=(-\infty,1)\): the supremum is \(1\) and it is not attained. The conditions (2.1) together with \(crQ=p-1<2x\) force \(cr<2x^{1-\theta}\). At \(\theta\ge1\) this contradicts \(cr\ge2\).
The exponent \(0.9\) of Theorem 6.1 of Primes with controlled predecessors is the cutoff in the weight (1.2) of Weights, mass and distribution in a progression. It gives the support bound \(r\le x^{0.11}\), and it guarantees that no odd prime \(j\le x^{0.002}\) equals \(Q\) in Lemma 4.1 of Primes with controlled predecessors. Theorem 2.1 keeps that cutoff and uses Theorem 6.1 only as a count; the gain comes from Lemma 1.1, applied after the count. Inserting \(r\le x^\varepsilon\) into the weight itself would break the group invariance of \(F\) that Lemma 2.1 of Weights, mass and distribution in a progression requires. This is why the condition is imposed by deletion.
Remark 4.2 (unstructured counts). OpenAI's preprint [PD, Theorem 1.1] states that the normalized logarithms of the prime factors of \(p-1\) follow the Poisson–Dirichlet law. For fixed \(\theta\in(0,1)\) it gives \[ \begin{aligned} &\#\{x<p\le2x:P^+(p-1)>p^\theta\}\\ &\quad\sim(1-\rho(1/\theta))\,\frac x{\log x}, \end{aligned} \] with \(\rho\) the Dickman function. Since \(\rho(u)=1-\log u\) on \([1,2]\), the proportion at \(\theta=0.9\) is \(-\log0.9=0.10536\ldots\). This count is larger than that of Theorem 2.1, but it carries none of the structure (2.1). This lesson does not use it.
5. Exercises
5.1. Show that \(\sum_{e^{L^{0.1}}<q<e^{L^{0.3}}}1/q=0.2\log L+o(1)\).
5.2. For \(\theta=0.99\), find \(\varepsilon\) in the proof of Theorem 2.1 and the exponent of \(x\) in the lower bound for \(Q/x^\theta\).
5.3. Show that for large \(x\) a prime \(p\in(x,2x)\) has at most one decomposition \(p-1=crQ\) with \(c\in\{2,4\}\) fixed, \(Q>x^{0.9}\) prime, and all prime factors of \(r\) below \(e^{L^{0.3}}\).
5.4. Apply Lemma 1.1 with \(T=L\), \(\mathcal Q\) the primes in \((e^2,x]\) and \(R=x^\varepsilon\), \(\varepsilon<1\). Show that it gives no saving, and explain why the upper end \(e^{L^{0.3}}\) of the range matters.
6. Solutions
5.1. By Mertens' estimate the sum is \(0.3\log L-0.1\log L+O(L^{-0.1})\). A possible prime at an endpoint changes it by \(o(1)\).
5.2. \(\varepsilon=\min(0.1,0.005)=0.005\), and \(Q/x^\theta>x^{1-\varepsilon-\theta}/8=x^{0.005}/8\).
5.3. Every prime factor of \(cr\) is below \(e^{L^{0.3}}<x^{0.9}\), so \(Q=P^+(p-1)\). Then \(r=(p-1)/(cQ)\) is determined.
5.4. Here \(R^{-1/T}=e^{-\varepsilon}\), while \(\exp(2e\sum_{e^2<q\le x}1/q)\) grows like \(L^{2e}\). So the bound (1.2) exceeds \(x/L^2\). With \(T=L^{0.3}\), the factor \(R^{-1/T}=e^{-\varepsilon L^{0.7}}\) beats every power of \(L\).
Sources
- [OpenAI-029] OpenAI, Primitive roots for every admissible integer base, OpenAI Math Release preprint, 4 October 2026, Proposition 2.1 and Section 12. https://github.com/openai/math/tree/main/preprints/Primitive-roots-for-every-admissible-integer-base-October-4-2026
- [PD] OpenAI, The Poisson–Dirichlet law for prime predecessors, OpenAI Math Release preprint, 24 September 2026, Theorem 1.1. https://github.com/openai/math/tree/main/preprints/The-Poisson-Dirichlet-Law-for-Prime-Predecessors-September-24-2026
- The main argument follows a solution produced with ChatGPT (OpenAI) in October 2026 and completed here.