Erdős Problem 346, limit-exists interpretation #
This file formalizes an attempted Lean 4 proof of the limit-exists interpretation of Erdős Problem 346.
Original problem: T. F. Bloom, Erdős Problem #346, https://www.erdosproblems.com/346
Background references:
- P. Erdős and R. L. Graham, Old and New Problems and Results in Combinatorial Number Theory, Monographie No. 28 de L'Enseignement Mathématique, 1980.
- R. L. Graham, "A Property of Fibonacci Numbers", Fibonacci Quarterly 2 (1964), 1-10.
Forum relation: In the Erdős Problems discussion thread for Problem #346, Liam Price submitted an answer for the convergence-from-hypotheses interpretation. Dogmachine then proposed the limit-exists interpretation, citing ambiguity in the Erdős-Graham phrasing and Graham's earlier formulation. This file formalizes that limit-exists interpretation:
if the two deletion hypotheses hold, the sequence has a uniform ratio gap, and
a (n+1) / a n tends to a real number L > 1, then the quotient sequence tends
to the golden ratio.
AI usage disclosure: The Lean formalization and accompanying writeup were prepared with assistance from Codex and ChatGPT. The final mathematical claims and public presentation remain the author's responsibility.
Completeness of the subsequence with index set I: every sufficiently large
natural number is a finite subset sum of terms indexed by I.
Equations
- Erdos346.IsCompleteOn a I = ∃ (H : ℕ), ∀ (n : ℕ), H ≤ n → n ∈ Erdos346.subsetSums a I
Instances For
The least threshold from which a complete subsequence is complete, once a proof of completeness is supplied.
Equations
- Erdos346.leastCompleteThreshold a I hcomplete = Nat.find hcomplete
Instances For
Finite subset sums using only terms in I up to index n.
Equations
- Erdos346.prefixSubsetSums a I n = Erdos346.subsetSums a (I ∩ Set.Iic n)
Instances For
The finite subset sums up to n cover the whole interval [H, T].
Equations
- Erdos346.CoversInterval a I n H T = ∀ (t : ℕ), H ≤ t → t ≤ T → t ∈ Erdos346.prefixSubsetSums a I n
Instances For
The least lower endpoint H from which the prefix through n covers
[H, T]. This is always defined: H = T + 1 gives a vacuous cover.
Equations
- Erdos346.leastPrefixCoverThreshold a I n T = Nat.find ⋯
Instances For
The least prefix-cover threshold really covers from itself.
Any prefix-cover threshold bounds the least such threshold.
Sum of the non-deleted terms up to index n.
Equations
- Erdos346.keptPrefixSum a D n = ∑ i ∈ Finset.range (n + 1) with i ∉ D, a i
Instances For
Sum of all terms up to index n.
Equations
- Erdos346.fullPrefixSum a n = ∑ i ∈ Finset.range (n + 1), a i
Instances For
The fixed numerical slack needed to convert a full-prefix surplus into the
Brown budget inequality for an initial covered interval [H, T] at index M.
Equations
- Erdos346.budgetBase a H M T = Erdos346.fullPrefixSum a M + H - (T + 1)
Instances For
A simple finite upper bound sufficient for all before-budget inequalities
between M and the first deleted index start. This is deliberately coarse:
it packages the finite obstruction into one number.
Equations
- Erdos346.beforeBudgetNeed a H start = (Finset.range start).sup fun (n : ℕ) => H + a (n + 1)
Instances For
Sum of the deleted terms up to index n.
Equations
- Erdos346.deletedPrefixSum a D n = ∑ i ∈ Finset.range (n + 1) with i ∈ D, a i
Instances For
Sum of the first k deleted terms in an indexed deletion sequence.
Equations
- Erdos346.indexedDeletedSum a b k = ∑ j ∈ Finset.range k, a (b j)
Instances For
State for the recursive sparse-deletion construction. The first component is the current deleted index, and the second component is the sum of all deleted terms chosen so far.
Equations
- One or more equations did not get rendered due to their size.
- Erdos346.recursiveBudgetState a threshold C start 0 = (start, a start)
Instances For
The indexed deletion sequence associated to recursiveBudgetState.
Equations
- Erdos346.recursiveBudgetSeq a threshold C start k = (Erdos346.recursiveBudgetState a threshold C start k).1
Instances For
Arithmetic-progression deletion indices. In the global-budget route these are used to delete one term every fixed block.
Equations
- Erdos346.periodicDeletionSeq start period k = start + k * period
Instances For
The set of indices deleted by the periodic deletion scheme with the given
start and period, i.e. the range of periodicDeletionSeq start period.
Equations
- Erdos346.periodicDeletionSet start period = Set.range (Erdos346.periodicDeletionSeq start period)
Instances For
Sum of the m terms immediately preceding index n.
For m ≤ n, this is a (n-m) + ... + a (n-1).
Equations
- Erdos346.previousBlockSum a m n = ∑ i ∈ Finset.Icc (n - m) (n - 1), a i
Instances For
From index N on, every term a (n + 1) is strictly dominated by the sum of
the previous m terms (previousBlockSum a m n). This is the fixed-length block
surplus property used to force completeness.
Equations
- Erdos346.FixedLengthBlockSurplusFrom a m N = ∀ (n : ℕ), N ≤ n → m ≤ n → a (n + 1) < Erdos346.previousBlockSum a m n
Instances For
The geometric block-surplus condition at ratio ρ and block length m:
for every n ≥ m, the inverse geometric weights of the previous m terms sum to
more than one. This drives the surplus needed for the golden-ratio bound.
Equations
Instances For
The quotient a (n+1) / a n, viewed in ℝ.
Equations
- Erdos346.quotient a n = ↑(a (n + 1)) / ↑(a n)
Instances For
The original lacunarity lower bound, expressed for index sequences.
Equations
- Erdos346.HasUniformRatioGap a = ∃ (ε : ℝ), 0 < ε ∧ ∀ (n : ℕ), 1 + ε ≤ Erdos346.quotient a n
Instances For
Deleting any finite set of indices leaves a complete subsequence.
Equations
- Erdos346.FiniteDeletionComplete a = ∀ (D : Set ℕ), D.Finite → Erdos346.IsCompleteOn a Dᶜ
Instances For
Uniform finite-base singleton completeness. This is the remaining uniformity
needed by the finite-initial-tail lower-bound route: after deleting one fixed
finite base set F, deleting any sufficiently late single extra index still
leaves a complete subsequence with one common threshold.
Equations
Instances For
The weaker uniform condition actually needed by the start = M + 2
finite-initial-tail seed route. Instead of asking for full completeness after
deleting F ∪ {M + 1}, it asks only for the finite prefix cover window used by
the seed constructor.
Equations
Instances For
A still weaker seed-window condition: the threshold may depend on M, but
for eventually all M there is a prefix-cover threshold small enough for the
two finite numeric inequalities used by the start = M + 2 seed constructor.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The seed construction only needs one sufficiently late good singleton window, not a good window at every sufficiently late index. This formulation asks for good finite-singleton windows arbitrarily far out.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The same small-window target phrased using the least possible prefix-cover threshold. This turns the remaining finite-deletion problem into bounding one explicit finite combinatorial quantity.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Least-threshold form of
ArbitrarilyLargeFinSingletonSmallPrefixCover.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The weakest current finite-window target for the lower-bound route. It
only asks, arbitrarily far out, for the first finite seed inequality. Under the
eventual upper ratio bound quotient a n < ρ < φ, finite-deletion completeness
upgrades this to the full least-small target below.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equivalent cover form of the current first-seed target. Instead of
speaking about the canonical least threshold, it asks directly that the finite
prefix through M cover the whole first seed window
[nextSeedMargin a M, a (M + 2) - 1].
Equations
- One or more equations did not get rendered due to their size.
Instances For
Local closure condition that rules out the first-window shift gap. Every
prefix sum below the first seed margin remains a prefix sum after adding
a (M + 1), even though index M + 1 itself is not allowed in the prefix
through M.
Equations
- Erdos346.FirstWindowShiftClosed a F M = ∀ s ∈ Erdos346.prefixSubsetSums a Fᶜ M, s < Erdos346.nextSeedMargin a M → a (M + 1) + s ∈ Erdos346.prefixSubsetSums a Fᶜ M
Instances For
Arbitrarily late occurrences of the local first-window shift closure for one fixed finite base set.
Equations
Instances For
The moving singleton-deletion threshold condition sufficient for
FirstWindowShiftClosed: after deleting F ∪ {M + 1}, completeness has already
started by the previous term a (M + 1).
Equations
Instances For
Arbitrarily late occurrences of the moving singleton-deletion threshold condition for one fixed finite base set.
Equations
Instances For
Deleting any infinite set of indices leaves an incomplete subsequence.
Equations
- Erdos346.InfiniteDeletionIncomplete a = ∀ (D : Set ℕ), D.Infinite → ¬Erdos346.IsCompleteOn a Dᶜ
Instances For
Completeness after deleting a set of values rather than a set of indices.
The remaining terms are those indices i whose value a i is not in B.
Equations
- Erdos346.IsCompleteAfterDeletingValues a B = Erdos346.IsCompleteOn a {i : ℕ | a i ∉ B}
Instances For
Deleting any finite set of values leaves a complete subsequence.
Equations
- Erdos346.ValueFiniteDeletionComplete a = ∀ (B : Set ℕ), B.Finite → Erdos346.IsCompleteAfterDeletingValues a B
Instances For
Deleting any infinite subset of the sequence values leaves an incomplete
subsequence. The side condition B ⊆ Set.range a says that B really consists
of values from the sequence.
Equations
- Erdos346.ValueInfiniteDeletionIncomplete a = ∀ (B : Set ℕ), B.Infinite → B ⊆ Set.range a → ¬Erdos346.IsCompleteAfterDeletingValues a B
Instances For
The canonical completeness threshold obtained from finite deletion after
deleting a fixed finite base set F and the moving singleton M + 1.
Equations
- Erdos346.finSingletonCompleteThreshold a hfinite hF M = Erdos346.leastCompleteThreshold a (F ∪ {M + 1})ᶜ ⋯
Instances For
A moving hole in a tail. Starting from a N - 1, it is propagated two
steps at a time by adding the intervening odd-indexed tail term.
Equations
- Erdos346.propagatedGap a N 0 = a N - 1
- Erdos346.propagatedGap a N k.succ = Erdos346.propagatedGap a N k + a (N + 2 * k + 1)
Instances For
Brown-style data guaranteeing that deleting D preserves eventual
completeness. The initial interval [H, keptPrefixSum M] is covered, every
kept next term is small enough to extend the covered interval, and the kept
prefix sums are unbounded.
- infinite : D.Infinite
- initial_cover : CoversInterval a Dᶜ M H (keptPrefixSum a D M)
Instances For
Brown-style data before proving unboundedness of the kept prefix sums.
Coinfiniteness of the kept set plus positivity implies the unbounded field of
BrownDeletionData.
- infinite : D.Infinite
- initial_cover : CoversInterval a Dᶜ M H (keptPrefixSum a D M)
Instances For
Brown-style seed data with a fixed initial slack C. This is weaker than
BrownStepData: the propagated covered interval is allowed to be
[H, C + keptPrefixSum n] rather than [H, keptPrefixSum n].
- infinite : D.Infinite
- initial_cover : CoversInterval a Dᶜ M H (C + keptPrefixSum a D M)
Instances For
Brown-style tail data when the initial cover has already been proved in
the kept set. This removes the avoids_initial requirement and is useful for
starting after a finite batch of deletions.
- infinite : D.Infinite
- initial_cover : CoversInterval a Dᶜ M H T
Instances For
A purely numerical sufficient condition for BrownTailData. It controls
the total deleted mass up to n, rather than mentioning the kept prefix sum in
the step inequality.
- infinite : D.Infinite
- budget (n : ℕ) : M ≤ n → n + 1 ∉ D → deletedPrefixSum a D n + a (n + 1) ≤ T + (fullPrefixSum a n - fullPrefixSum a M) - H + 1
Instances For
Numerical sufficient condition for BrownTailCoverData. Unlike
BrownBudgetData, this permits a finite batch of deletions at or before M,
provided the initial cover is already in the kept set.
- infinite : D.Infinite
- initial_cover : CoversInterval a Dᶜ M H T
- budget (n : ℕ) : M ≤ n → n + 1 ∉ D → deletedPrefixSum a D n + a (n + 1) + fullPrefixSum a M + H ≤ T + fullPrefixSum a n + deletedPrefixSum a D M + 1
Instances For
Indexed sparse-tail data after a finite initial deletion set F. The
initial interval is covered in Fᶜ; the indexed tail starts after M, so the
initial cover also lives in the full kept set (F ∪ range b)ᶜ.
- initial_cover : CoversInterval a Fᶜ M H T
- budget (n : ℕ) : M ≤ n → n + 1 ∉ initialTailDeletionSet F b → deletedPrefixSum a (initialTailDeletionSet F b) n + a (n + 1) + fullPrefixSum a M + H ≤ T + fullPrefixSum a n + deletedPrefixSum a (initialTailDeletionSet F b) M + 1
Instances For
Constructive version of IndexedBudgetInitialTailData. It only asks for
the budget inequality before the first tail deletion and between consecutive
tail deletions.
- initial_cover : CoversInterval a Fᶜ M H T
- budget_before (n : ℕ) : M ≤ n → n < b 0 → n + 1 ∉ initialTailDeletionSet F b → deletedPrefixSum a (initialTailDeletionSet F b) n + a (n + 1) + fullPrefixSum a M + H ≤ T + fullPrefixSum a n + deletedPrefixSum a (initialTailDeletionSet F b) M + 1
- budget_between (k n : ℕ) : b k ≤ n → n < b (k + 1) → n + 1 ∉ initialTailDeletionSet F b → indexedDeletedSum a b (k + 1) + a (n + 1) + fullPrefixSum a M + H ≤ T + fullPrefixSum a n + 1
Instances For
Indexed sparse-deletion budget data. Unlike the discarded block-control condition below, this asks for the Brown budget inequality itself at the kept extension steps.
- budget (n : ℕ) : M ≤ n → n + 1 ∉ Set.range b → deletedPrefixSum a (Set.range b) n + a (n + 1) ≤ T + (fullPrefixSum a n - fullPrefixSum a M) - H + 1
Instances For
A more constructive form of indexed sparse-deletion budget data. It only
mentions the finite sum of the first k chosen deleted terms on the interval
before the next deletion.
Instances For
Finite seed data for the remaining recursive sparse-deletion construction.
After start, the global pair-prefix surplus handles all later intervals; the
finite interval before start is recorded explicitly.
- initial_cover : CoversInterval a Set.univ M H T
- surplus_from (n : ℕ) : start ≤ n → budgetBase a H M T + a n + a (n + 1) ≤ fullPrefixSum a n
Instances For
Seed data for the finite-initial-deletion route. The initial covered
interval is already proved after deleting a finite set F ⊆ Iic M; after
start, the usual pair-prefix surplus controls the recursive sparse tail.
- initial_cover : CoversInterval a Fᶜ M H T
- surplus_from (n : ℕ) : start ≤ n → budgetBase a H M T + a n + a (n + 1) ≤ fullPrefixSum a n
Instances For
The explicit finite/tail conditions that are sufficient to produce an
IndexedPartialBudgetSeed. This packages the remaining fixed-point choice:
find a prefix cover whose upper endpoint is large enough for the finite
before-budget, and whose budgetBase surplus starts before start.
- cover : CoversInterval a Set.univ M H T
- tail_surplus (n : ℕ) : start ≤ n → budgetBase a H M T + a n + a (n + 1) ≤ fullPrefixSum a n
Instances For
Block version of the remaining deletion-control data. Blocks allow the construction to delete a term and wait until the current term is large enough before the next kept-extension step.
- deleted_bound (n : ℕ) : M ≤ n → n + 1 ∉ blockDeletionSet b e → deletedPrefixSum a (blockDeletionSet b e) n ≤ a n
Instances For
The earlier block-control condition is too strong: at the end of the second deleted block the deleted prefix already contains a positive earlier deleted term plus the current block endpoint.
Proved analytic steps #
The first theorem below turns L > φ into the eventual two-step gap. The
gap-propagation theorem after it is also proved in Lean. The lower-bound
L < φ side is handled later by the axiom-free global-budget periodic deletion
construction.
If the ratio limit is larger than the golden ratio, then eventually there is a two-step gap.
A tail satisfying the eventual two-step gap cannot be complete.
The Brown-style sparse deletion data gives the lower-bound deletion counterexample needed in the main proof.
If the ratio limit lies strictly between 1 and φ, then one can delete
an infinite, sufficiently sparse set of indices and still remain complete. The
limit-to-eventual-bound part is proved here; the remaining sparse-deletion
construction is isolated above.
Consequences of the inputs #
Conditional value theorem for the interpretation where the ratio limit is assumed to exist and to be greater than one.
Legacy conditional value theorem under the explicit uniform finite-base singleton completeness condition. The main theorem no longer needs this route, but it remains as a named sufficient condition for the older singleton-threshold approach.
A fully proved variant of the value theorem under the weaker uniform prefix cover condition used by the finite-initial-tail seed.
Value theorem under the small, M-dependent prefix-cover window condition.
This is weaker than the uniform prefix-cover condition: the threshold can vary
with M, provided it is small enough for that M.
Value theorem under the still weaker condition that suitable small finite-singleton windows occur arbitrarily far out.
Value theorem under the current weakest finite-window condition: arbitrarily far out, only the first seed inequality is assumed.
Value theorem under the equivalent direct-cover form of the current weakest finite-window condition.
Same theorem phrased as convergence to the golden ratio.
Convergence form of
ratio_limit_eq_goldenRatio_of_uniformFinSingletonComplete.
Convergence form of
ratio_limit_eq_goldenRatio_of_uniformFinSingletonPrefixCover.
Convergence form of
ratio_limit_eq_goldenRatio_of_eventualFinSingletonSmallPrefixCover.
Convergence form of
ratio_limit_eq_goldenRatio_of_arbLargeFinSingletonSmallPrefixCover.
Convergence form of
ratio_limit_eq_goldenRatio_of_arbLargeFinSingletonFirstSmallPrefixCover.
Convergence form of
ratio_limit_eq_goldenRatio_of_arbLargeFinSingletonFirstPrefixCover.
Compact formulation: if a ratio limit L > 1 exists, then it is φ.
Compact formulation of the axiom-free conditional route using the explicit uniform finite-base singleton completeness hypothesis.
Compact formulation of the axiom-free conditional route using only the weaker uniform finite-base prefix-cover hypothesis.
Compact formulation of the axiom-free conditional route using the
M-dependent small prefix-cover window condition.
Compact formulation using only arbitrarily late good finite-singleton prefix-cover windows.
Compact formulation using only arbitrarily late first seed inequalities.
Compact formulation using the direct finite-prefix cover of the first seed
window [nextSeedMargin a M, a (M + 2) - 1].
Compact formulation using the local first-window shift-closure condition.
Finite-deletion completeness supplies the prefix cover through M + 1; the
shift-closure condition is exactly what lowers that cover to the prefix through
M.
Compact formulation using the moving singleton-deletion threshold condition:
arbitrarily far out, after deleting F ∪ {M + 1}, completeness starts no later
than a (M + 1).
Compact formulation using the least-threshold phrasing of the small finite-window condition.
Compact formulation using the least-threshold phrasing of the arbitrarily late small finite-window condition.
Short public-facing name for the main theorem.
Expanded statement for human inspection.
This is definitionally the same theorem as main, but the three compact
hypotheses HasUniformRatioGap, FiniteDeletionComplete, and
InfiniteDeletionIncomplete are expanded in the theorem statement.
Public-facing theorem using value deletion rather than index deletion.
Finite and infinite deletion hypotheses are stated for sets B : Set ℕ of
sequence values.
Expanded value-deletion statement for human inspection.