Proved theorem Unrefereed

Erdős–Straus: a sieve bound on the Type II branch

No dated manuscript catalogued; standalone verification archive linked above. On this site since · last revised · Verify & co-author
Kind
Theorem with a written proof (a large-sieve argument)
Status
Proof read by a falsifier agent with no gap found; unrefereed
Prior work
Not found in the sources searched; the method is standard, and the theorem is a transposition of Dahan's fixed-depth theorem to fixed shifts
On this page
  1. The statement
  2. Check it yourself
  3. Worked examples
  4. Status and who checked it
  5. Prior work
  6. Further reading

The statement

The Erdős–Straus conjecture says that 4/n=1/x+1/y+1/z4/n = 1/x + 1/y + 1/z has a solution in positive integers for every n≥2n \ge 2. For a prime p≡3(mod4)p \equiv 3 \pmod 4 the least shift is 1; the theorem concerns p≡1(mod4)p \equiv 1 \pmod 4. A shift c≡3(mod4)c \equiv 3 \pmod 4 proposes the denominator xc=(p+c)/4x_c = (p + c)/4; the Type II branch of shift cc succeeds when there are coprime d1,d2d_1,\allowbreak d_2 with d1d2∣xcd_1 d_2 \mid x_c and c∣d1+d2c \mid d_1 + d_2, and then gives a solution with that xx. Let k(p)k(p) be the least shift that works on either branch.

Theorem F. Let CC be a finite set of positive integers c≡3(mod4)c \equiv 3 \pmod 4. The number of primes p≤Np \le N, p≡1(mod4)p \equiv 1 \pmod 4, at which every c∈Cc \in C fails on the Type II branch is

≪C  N(log⁡N)−1−∣C∣/2.\ll_C\; N (\log N)^{-1-|C|/2}.

Corollary. For K≥3K \ge 3 let m=⌊(K+1)/4⌋m = \lfloor (K + 1)/4 \rfloor, the number of shifts 3, 7, 11, … up to KK. Then #{p≤N:k(p)>K}≪KN(log⁡N)−1−m/2\#\{p \le N : k(p) > K\} \ll_K N (\log N)^{-1-m/2}. At K=7K = 7: #{p≤N:k(p)>7}≪N(log⁡N)−2\#\{p \le N : k(p) > 7\} \ll N (\log N)^{-2}.

Check it yourself

  • One shift at work. For p=3361p = 3361, x3=(3361+3)/4=841=292x_3 = (3361 + 3)/4 = 841 = 29^2, and 29 leaves remainder 2 on division by 3. Take d1=1d_1 = 1, d2=29d_2 = 29: they are coprime, d1d2∣841d_1 d_2 \mid 841, and 3 divides 1+291 + 29. So k(3361)=3k(3361) = 3.
  • The key step (the involution lemma). If shift cc fails on the Type II branch at pp, the residues mod cc of the prime factors of xcx_c contain no pair a,−aa,\allowbreak -a: two primes q1≡aq_1 \equiv a and q2≡−aq_2 \equiv -a dividing xcx_c would give d1=q1d_1 = q_1, d2=q2d_2 = q_2. So each failing shift confines the prime factors of xcx_c to half the residue classes coprime to cc, and each such shift removes about half a class per prime from the sieve.
  • The lemma on data. Over the primes p≡1(mod4)p \equiv 1 \pmod 4 below 1048576 and the shifts 3, 7, 11, 15, 19, 23: 81307 Type II failures and 0 violations of the lemma, while its converse fails 95454 times. The converse fails because products also count: 2⋅3≡−1(mod7)2 \cdot 3 \equiv -1 \pmod 7.

Worked examples

The counts of primes p≡1(mod4)p \equiv 1 \pmod 4 up to NN at which the Type II branch fails, with the ratio to the theorem's rate in brackets:

NN fails at 3 / N(log⁡N)−1.5N(\log N)^{-1.5} at 3 and 7 / N(log⁡N)−2N(\log N)^{-2} at 3, 7, 11 / N(log⁡N)−2.5N(\log N)^{-2.5}
2162^{16} 418 (0.2356) 88 (0.1652) 33 (0.2063)
2182^{18} 1,401 (0.2355) 301 (0.1787) 119 (0.2496)
2202^{20} 4,742 (0.2334) 1,033 (0.1893) 420 (0.2866)
2222^{22} 16,407 (0.2329) 3,374 (0.1871) 1,329 (0.2877)
2242^{24} 57,186 (0.2313) 11,270 (0.1859) 4,196 (0.2823)

The middle column equals #{p≤N:k(p)>7}\#\{p \le N : k(p) > 7\} at every NN. The third column counts Type II failures, not k(p)>11k(p) > 11. The ratios settle, which agrees with the theorem's exponent but cannot confirm it: the bound is an upper bound and these NN are small.

Status and who checked it

  • Proved by hand. The proof combines the involution lemma with Montgomery's arithmetic large sieve at Q=NQ = \sqrt N and Mertens' theorem in arithmetic progressions, both cited as standard results. No computation is part of the proof.
  • Who checked it. A falsifier agent read the proof step by step and found no gap ("this is a reading, not an independent proof"), and reran the checks above. No second reviewer graded it. Unrefereed.
  • Open. A matching lower bound for k(p)>7k(p) > 7 without a hypothesis: it is not even known from these sources that infinitely many primes have k(p)>7k(p) > 7. Under the Elliott–Halberstam conjecture a lower bound of the same order holds, by a conditional argument the same falsifier agent also held. Whether the exponent 1+∣C∣/21 + |C|/2 is sharp for two or more shifts.

Prior work

Credited. The criterion at a fixed shift, its two branches and the shift-3 case are in Dahan's work (arXiv 2608.24035), as is the nearest theorem: "fixed_depth", the same shape of bound for fixed sets of witness pairs whose shift moves with nn, by Selberg's sieve and Bombieri–Vinogradov. Dahan's later preprint (arXiv 2611.08717) applies the large sieve over a growing number of shifts, with a weaker saving at any fixed number. Shift 3 alone has a two-sided count by Iwaniec (1972); shift 7 alone by Dahan.

Searches run. The sources listed for the project, and Dahan's two papers searched for this statement. The theorem was not found in the sources searched. The method is standard; the theorem is a transposition of fixed_depth to fixed shifts.

Still owed. The body of Sander (1994), which counts one family and would not decide this; a wider search than the sources listed.

Further reading