erdos-problems 0.3.1 → 0.3.3

package/packs/number-theory/problems/848/ROUTE_PACKET.yaml

@@ -0,0 +1,16 @@
+route_packet_id: nt848_finite_check_gap_closure_v1
+route_id: finite_check_gap_closure
+frontier_claim: Close the finite decidable gap for Problem 848 while keeping imported threshold progress separate from repo-owned audited claims.
+theorem_module: ""
+checkpoint_packet: CHECKPOINT_TEMPLATE.md
+report_packet: REPORT_TEMPLATE.md
+ready_prompts:
+  - What is the best imported explicit `N0` currently visible on the public thread, and how trusted is it?
+  - What exact interval should the repo try to certify first in the bounded-verification lane?
+  - What certificate format is required before a claimed interval counts as canonically covered?
+  - Which public verification attempts should be treated as audit inputs rather than accepted coverage?
+  - How should the finite remainder be partitioned into regimes so that methods and trust levels stay visible?
+verification_hook:
+  - erdos number-theory status 848
+  - erdos number-theory routes 848
+  - erdos problem show 848

package/packs/number-theory/problems/848/THEOREM_STYLE_EXPLICIT_NOTE.md

@@ -0,0 +1,91 @@
+# Problem 848 Theorem-Style Explicit Note
+
+This note closes `N848.G1.A12`.
+
+## Statement
+
+**Repo explicit candidate.**
+
+Let
+- `N >= exp(1420)`,
+- `A subseteq [N]`,
+
+and suppose that
+- `ab + 1` is never squarefree for all `a, b in A`,
+- `|A| >= (1/25 - 10^-4) * N`.
+
+Then the current repo candidate says:
+- `A` is contained in either the `7 mod 25` class or the `18 mod 25` class.
+
+## Witness parameters
+
+The shared witness used throughout the current repo candidate is:
+- truncation value `T = 250`
+- threshold scale `N >= exp(1420)`
+- stability parameter `eta = 10^-4`
+
+## Proof skeleton
+
+### Step 1. Split into the public branches from Sawhney's proof
+
+The public proof breaks into the branches whose rounded bounds are:
+- `0.0377`
+- `0.0358`
+- `0.0336`
+- `0.0294`
+
+### Step 2. Use the shared explicit witness
+
+The repo now has explicit ledger support for the shared witness:
+- `LEMMA21_EXPLICIT_BOUND.md`
+- `LEMMA21_TRUNCATION_SCAN.md`
+- `LEMMA22_EXPLICIT_BOUND.md`
+- `LEMMA22_PRIME_COUNT_BOUND.md`
+- `CERTIFIED_NUMERICAL_LEDGER.md`
+
+These notes provide the explicit replacements for the original `o(1)` and `<<` steps at the
+chosen witness scale.
+
+### Step 3. Close the tightest branch
+
+`WEAKEST_BRANCH_T250_ASSEMBLY.md` shows that the tightest public branch retains a certified
+visible reserve of about
+- `2.66e-4`
+
+after all currently frozen lemma-level costs and the working choice
+- `eta = 10^-4`.
+
+### Step 4. Check the remaining public branches
+
+`BRANCH_COMPARISON_LEDGER.md` shows that the branches `0.0358`, `0.0336`, and `0.0294` all
+retain strictly larger visible reserve than the weakest branch under the same witness.
+
+So no new branch obstruction appears at the current explicit level.
+
+### Step 5. Conclude the repo candidate
+
+Since every public branch in the proof skeleton is controlled by the same witness, the repo
+arrives at the proposition-level candidate stated above.
+
+## What this note is and is not
+
+This note **is**:
+- a theorem-shaped packaging of the current repo candidate
+- a bridge artifact for paper-writer mode
+- a clean summary of the current explicit witness
+
+This note is **not yet**:
+- a final publication-ready proof
+- a formal proof-assistant certificate
+- a declaration that Problem 848 is fully solved in the repo
+
+## Remaining gap
+
+The remaining work is no longer “find the right witness.”
+
+It is:
+- decide whether the current repo candidate is strong enough to publish internally as a public
+  review artifact
+- port it into the paper bundle
+- and, if desired, replace the current repo-level interval reasoning with stronger formal or
+  machine-checked certification

package/packs/number-theory/problems/848/THRESHOLD_LEDGER.md

@@ -0,0 +1,132 @@
+# Problem 848 Threshold Ledger
+
+This file records the exact shape of the remaining finite-check gap for Problem 848.
+
+It is intentionally conservative:
+- do not promote tentative forum bounds to canonical theorem claims
+- separate existential threshold statements from explicit ones
+- separate sample finite checks from full finite closure
+
+## Current route claim
+
+- Publicly established: there exists an integer `N0` such that the desired extremal statement holds for all `N >= N0`.
+- Not yet canonically established here: a complete explicit threshold `N0`, or a finished proof for every `N`.
+
+## Source-backed threshold map
+
+### 1. Proposition 1.1 in Sawhney's note
+
+Source:
+- Sawhney, `Problem_848.pdf`, page 1
+
+Public statement:
+- There exists an integer `N0` such that for all `N >= N0`, if `A ⊆ [N]` and `ab+1` is never squarefree for all `a,b in A`, then
+  `|A| <= |{n in [N] : n ≡ 7 mod 25}|`.
+- Equality is achieved by the `7 mod 25` class and possibly the `18 mod 25` class.
+
+Classification:
+- `existential`
+
+Why:
+- The theorem statement is explicitly existential.
+- The note does not state a usable numerical `N0`.
+
+### 2. Lemma 2.1
+
+Source:
+- Sawhney, `Problem_848.pdf`, page 1
+
+Public shape:
+- The lemma gives an asymptotic count with error `<< N (log N)^(-1/2)`.
+- The proof also uses an `N^(o(1))` style term through the product over small primes.
+
+Classification:
+- `weakly explicit`
+
+Why:
+- The proof architecture is explicit enough in principle to support numerical extraction.
+- The constants are not packaged in the note as a direct computational threshold.
+
+### 3. Lemma 2.2
+
+Source:
+- Sawhney, `Problem_848.pdf`, pages 1-2
+
+Public shape:
+- The lemma gives an asymptotic estimate for squarefree obstructions on arithmetic progressions with error `<< N / sqrt(log N)`.
+
+Classification:
+- `weakly explicit`
+
+Why:
+- The structure is explicit in principle.
+- The forum discussion repeatedly identifies this lemma as the harder place to sharpen if one wants a realistic explicit threshold.
+
+### 4. Final casework numerics in the proof
+
+Source:
+- Sawhney, `Problem_848.pdf`, pages 2-4
+
+Public shape:
+- The proof records rounded numerical bounds such as `0.0377`, `0.0358`, `0.0336`, and `0.0294`.
+- These are used after absorbing `o(1)` errors under the assumption that `N` is sufficiently large.
+
+Classification:
+- `explicit case numerics + existential threshold absorption`
+
+Why:
+- The local case splits and main-term constants are concrete.
+- The point where the error terms become small enough is still not canonically explicit in the note.
+
+## Public threshold-improvement chatter
+
+These are useful research leads, not canonical solved artifacts.
+
+### 5. Forum-level explicit thresholds
+
+Source:
+- `erdosproblems.com/forum/thread/848`
+
+Publicly mentioned:
+- `exp(1958)` from a GPT-assisted threshold-extraction effort discussed by Terence Tao
+- `exp(1420)` as a later plausible improvement discussed on the same thread
+
+Classification:
+- `tentative / not canonically frozen here`
+
+Why:
+- These values appear in forum discussion, not in the public theorem statement on the main problem page.
+- The thread itself treats them as intermediate and in need of verification or better writeup.
+
+## Public finite-verification coverage
+
+### 6. Finite checks discussed publicly
+
+Source:
+- `erdosproblems.com/forum/thread/848`
+
+Publicly mentioned:
+- straightforward approaches only checked the problem up to a few hundred in one discussion branch
+- the public Lean thread claims finite verification for `N = 50` and `N = 100`
+
+Classification:
+- `partial finite coverage`
+
+Why:
+- This is not a complete all-`N` finite closure.
+- It does, however, show that some finite verification machinery already exists in public.
+
+## Immediate implications
+
+- The cleanest honest next step is to turn the existential/weakly-explicit split above into a line-by-line extraction checklist from Sawhney's note.
+- The weakest branch is now isolated in `WEAKEST_CASE_BUDGET.md`, including the exact place where the proof spends rounded branch numerics without yet freezing the usable explicit slack.
+- The repo now also has a conservative numerical freeze of that weakest branch, suggesting the real main-term branch bound is about `0.0376113` with slack about `0.0023887` before analytic error absorption.
+- The second step is to decide whether the best near-term route is:
+  - improving Lemma 2.2 and related error terms enough to get a practical explicit `N0`, or
+  - treating the current explicit-threshold discussion as support context and investing in bounded finite computation.
+
+## What not to claim yet
+
+- Do not claim a canonical explicit threshold `N0`.
+- Do not claim the forum bounds `exp(1958)` or `exp(1420)` are fully verified repo truth.
+- Do not claim the public Lean files already close the entire finite remainder.

package/packs/number-theory/problems/848/VERIFICATION_CERTIFICATE_SPEC.md

@@ -0,0 +1,60 @@
+# Problem 848 Verification Certificate Spec
+
+Any bounded-verification claim for Problem `848` should attach a certificate with these fields.
+
+## Required fields
+
+- `interval`
+  - exact range covered, for example `N = 1..7306`
+- `method_class`
+  - one of:
+    - `exact_small_n`
+    - `structured_breakpoints`
+    - `imported_external_computation`
+    - `hybrid`
+- `claim_level`
+  - one of:
+    - `exact`
+    - `verified`
+    - `external_only`
+- `artifacts`
+  - concrete files, repos, scripts, or notes used for the claim
+- `reproduction`
+  - how a maintainer can rerun or recheck the claim
+- `failure_mode`
+  - what would invalidate the interval claim
+- `audit_status`
+  - one of:
+    - `repo_audited`
+    - `partially_audited`
+    - `external_unreviewed`
+    - `blocked`
+
+## Strongly preferred fields
+
+- `breakpoint_definition`
+  - if the method uses breakpoint sufficiency, define exactly what the breakpoints are
+- `monotonicity_justification`
+  - if interval coverage is inferred between checkpoints, explain why
+- `independent_check`
+  - a second script, proof note, or reviewer confirmation
+- `cost_profile`
+  - rough runtime or proof complexity
+
+## Rejection triggers
+
+Do not count a range as canonically covered if any of the following is true:
+
+- the claim silently uses the wrong asymptotic handoff threshold
+- the method covers only sample points but is presented as interval coverage
+- a breakpoint argument is used without a monotonicity justification
+- the computation is public but explicitly criticized as unverified or likely incorrect and no
+  repo audit has answered that criticism
+- the result cannot be rerun or inspected by a future maintainer
+
+## Why this matters here
+
+The public thread already contains one verification attempt that was later corrected and
+criticized as difficult to verify. That does not make bounded verification hopeless. It does
+mean this repo should demand interval certificates rather than encouraging "I think this
+covers everything up to X" style claims.

package/packs/number-theory/problems/848/VERIFICATION_REGIMES.md

@@ -0,0 +1,87 @@
+# Problem 848 Verification Regimes
+
+This file partitions the finite remainder into regimes so the repo can track progress
+without pretending that one method fits the whole gap.
+
+## Trusted public asymptotic handoff
+
+- public theorem: Sawhney proves the statement for all sufficiently large `N`
+- imported explicit threshold timeline:
+  - `7 x 10^17` on 2026-03-21
+  - `3.3 x 10^17` on 2026-03-22
+  - `2.64 x 10^17` on 2026-03-23
+- current repo posture:
+  - these thresholds are imported public claims
+  - they size the finite remainder operationally
+  - they are not yet promoted to canonical repo-owned theorem statements
+
+## Verification regime split
+
+### Regime A: exact small-`N` coverage
+
+Goal:
+- cover a small base interval by exact methods that can be rerun and independently checked
+
+Candidate methods:
+- direct search
+- pair-exchange lemmas
+- exact residue-structure enumeration
+- very small witness-prime case splits
+
+Desired certificate:
+- exact witness/counterexample status for each `N`
+- reproducible script or proof note
+
+### Regime B: structured breakpoint verification
+
+Goal:
+- cover a medium interval by checking a monotone or breakpoint-sufficient inequality rather
+  than every raw instance independently
+
+Candidate methods:
+- structural breakpoint tables
+- clique/exchange inequalities
+- monotonicity-ledger arguments
+
+Desired certificate:
+- exact breakpoint set definition
+- proof that breakpoint coverage implies interval coverage
+- machine-readable result table
+
+### Regime C: audited imported computation
+
+Goal:
+- record public computational claims without treating them as canonical until the repo has an
+  audit story
+
+Candidate inputs:
+- public repos
+- public verification notes
+- independently rerun scripts
+
+Desired certificate:
+- imported source URL
+- exact claimed interval
+- audit outcome: accepted, external-only, or blocked
+
+### Regime D: asymptotic handoff
+
+Goal:
+- state the point above which the public theorem takes over
+
+Desired certificate:
+- cited theorem source
+- exact threshold assumption being used operationally
+- note whether the threshold is imported-only or repo-audited
+
+## Current honest posture
+
+- Regime A: not frozen in this repo
+- Regime B: not frozen in this repo
+- Regime C: public attempts exist, but some were explicitly criticized on the forum as
+  difficult to verify or likely incorrect
+- Regime D: public asymptotic theorem exists; imported explicit thresholds are tracked but not
+  yet repo-audited
+
+So the next honest move is to freeze the certificate format before claiming any new verified
+interval.

package/packs/number-theory/problems/848/WEAKEST_BRANCH_T250_ASSEMBLY.md

@@ -0,0 +1,109 @@
+# Problem 848 Weakest Branch Assembly at `T = 250`
+
+This note closes `N848.G1.A7`.
+
+Goal:
+- turn the `T = 250` weakest-branch witness budget into a line-by-line branch ledger
+- freeze an honest working `eta` value for this branch
+
+Scope:
+- this is only the weakest public `0.0377` branch
+- this is not yet a full explicit-threshold proof for Proposition 1.1
+- the remaining repo question is whether the same witness closes the other public branches
+
+## Branch setup
+
+Assume:
+- `N >= exp(1420)`
+- `eta > 0`
+- `|A| >= (1/25 - eta) * N`
+- the branch with an even element `b in A* = A \\ (A7 union A18)` is active
+- truncation witness `T = 250`
+
+## Line-by-line upper bound ledger
+
+### 1. `A*` main term
+
+From `WEAKEST_CASE_BUDGET.md`:
+- `A* / N <= 0.0251587645... + remainder`
+
+### 2. `A*` explicit remainder
+
+From `LEMMA21_TRUNCATION_SCAN.md`:
+- `A*` tail density at `T = 250`: about `0.0005641453`
+- `A*` discrete term: `23 * (3^23 - 1) / N`
+
+At `N >= exp(1420)`, the discrete term is astronomically small and far below the visible
+decimal budget.
+
+### 3. `A7 union A18` main term
+
+From `WEAKEST_CASE_BUDGET.md`:
+- `(A7 union A18) / N <= 0.0124525434... + remainder`
+
+### 4. `A7 union A18` explicit remainder
+
+From `LEMMA22_EXPLICIT_BOUND.md` and `LEMMA22_PRIME_COUNT_BOUND.md`:
+- Lemma 2.2 `N / p^2` tail density at `T = 250`: about `0.0000491054`
+- Lemma 2.2 prime-count term at `N >= exp(1420)`: at most `0.0014095427`
+- Lemma 2.2 discrete term: `2 * (2^51 - 1) / N`
+
+Again, the discrete term is astronomically small at `N >= exp(1420)`.
+
+## Combined explicit branch bound
+
+Collecting the frozen decimal contributions gives
+
+- branch main term: `0.0376113079`
+- `A*` tail: `0.0005641453`
+- Lemma 2.2 tail: `0.0000491054`
+- Lemma 2.2 prime-count term: `0.0014095427`
+
+So the visible explicit branch bound is
+
+- `0.0376113079 + 0.0005641453 + 0.0000491054 + 0.0014095427`
+- `= 0.0396341013` up to rounding in the displayed decimals
+
+This leaves visible room to `1/25 = 0.04` of about
+
+- `0.04 - 0.0396341013 = 0.0003658987`.
+
+## Working `eta` choice
+
+The branch contradiction only needs a positive margin after paying for `eta`.
+
+A conservative working choice is
+
+- `eta_work = 0.0001`.
+
+Then the remaining visible reserve is still about
+
+- `0.0003658987 - 0.0001 = 0.0002658987`.
+
+So with this working choice, the branch still retains about `2.66e-4` of visible room for
+the tiny discrete terms and any final rounding/bookkeeping losses.
+
+## Honest branch conclusion
+
+At the witness scale
+- `N >= exp(1420)`,
+- `T = 250`,
+- `eta = 10^-4`,
+
+the weakest public branch now appears explicitly contradictory.
+
+What this does **not** yet prove:
+- that `exp(1420)` is a fully verified threshold for the whole proposition
+- that the stronger public candidate thresholds are unnecessary
+- that every other public branch has already been checked with the same explicit witness
+
+What it **does** prove for the repo route:
+- the weakest public branch is no longer the blocker
+- the repo now has a concrete explicit branch witness with positive visible reserve
+
+## Next route consequence
+
+The next honest move is:
+- carry the same witness philosophy into the other public proof branches
+- check whether `T = 250`, `N >= exp(1420)`, and `eta = 10^-4` also close them
+- if one branch needs a different witness, record that precisely instead of forcing uniformity

package/packs/number-theory/problems/848/WEAKEST_BRANCH_T250_BUDGET.md

@@ -0,0 +1,107 @@
+# Problem 848 Weakest Branch Budget at `T = 250`
+
+This note closes `N848.G1.A6` at a witness-budget level.
+
+Goal:
+- test whether the weakest public `0.0377` branch is still blocked once both Lemma 2.1 and
+  Lemma 2.2 use a larger witness truncation parameter
+
+Witness choice:
+- `T = 250`
+
+Reason for this choice:
+- it is large enough to shrink the live `A*` tail materially
+- it is still tiny compared with the public candidate scales `exp(1420)` and `exp(1958)`
+- the discrete inclusion-exclusion terms remain negligible at those scales
+
+## Frozen main-term branch data
+
+From `WEAKEST_CASE_BUDGET.md`:
+- `A*` main term: about `0.0251587645`
+- `A7 union A18` main term: about `0.0124525434`
+- combined weakest-branch main term: about `0.0376113079`
+- branch slack to `1/25 = 0.04`: about `0.0023886921`
+
+## Explicit remainder ledger at `T = 250`
+
+### 1. `A*` via Lemma 2.1
+
+From `LEMMA21_TRUNCATION_SCAN.md`:
+- `A*` tail density at `T = 250`: about `0.0005641453`
+- `A*` discrete term: `23 * (3^23 - 1) / N`
+
+At `N = exp(1420)`:
+- the `A*` discrete density is below about `exp(-1392)`
+
+### 2. `A7 union A18` via Lemma 2.2
+
+From `LEMMA22_EXPLICIT_BOUND.md`:
+- Lemma 2.2 tail density at `T = 250`: about `0.0000491054`
+- worst-case Lemma 2.2 discrete term: `2 * (2^51 - 1) / N`
+- prime-count term from `LEMMA22_PRIME_COUNT_BOUND.md`: at most `0.0014095427` for
+  every `N >= exp(1420)`
+
+At `N = exp(1420)`:
+- the Lemma 2.2 discrete density is below about `exp(-1384)`
+
+### 3. Combined explicit remainder
+
+Adding the two live tail terms gives
+
+- combined tail density about `0.0006132507`
+
+and the combined discrete terms are still negligible at the public candidate scales.
+
+What is still missing:
+- no further lemma-level tail term; the remaining work is branch assembly and `eta`
+
+## Combined weakest-branch budget
+
+Main term plus the live explicit tails:
+
+- `0.0376113079 + 0.0006132507 = 0.0382245586`
+
+So the remaining room to `0.04` is still about
+
+- `0.0017754414`
+
+before the final branch bookkeeping.
+
+After inserting the explicit Lemma 2.2 prime-count bound at `N >= exp(1420)`,
+
+- `0.0017754414 - 0.0014095427 ~= 0.0003658987`
+
+So the currently visible room for
+- the final `eta` choice
+- and any remaining line-by-line bookkeeping losses
+
+is about `3.66e-4`.
+
+## Honest reading
+
+This does **not** yet prove an explicit threshold.
+
+What it **does** show is:
+- once `T` is allowed to move, the weakest branch no longer looks blocked by the two lemma
+  `N / p^2` tails themselves
+- the discrete inclusion-exclusion losses are no longer the issue
+- the Lemma 2.2 prime-count term also fits inside the branch budget at the public candidate
+  threshold scale
+- the route now has a concrete surviving witness margin of about `3.66e-4` for final
+  `eta` and bookkeeping losses
+
+So the live unresolved work is narrower than before:
+- write the line-by-line full branch assembly with the larger `T`
+- record the exact `eta` room that survives that assembly
+- then turn that into a genuine explicit-threshold witness candidate
+
+## Route consequence
+
+This means the current proof skeleton now looks plausibly explicitizable in the weakest
+branch at the witness scale `N >= exp(1420)`, provided the remaining line-by-line assembly
+does not consume more than about `3.66e-4` of density.
+
+That is a materially stronger posture than the one the repo started with:
+- the route is no longer waiting on a mystery bottleneck
+- it has a concrete witness truncation value
+- and it has a concrete surviving witness margin after all currently frozen lemma-level costs