{"id":"ea543c83-4df0-4963-8bc0-d073586b24df","arxiv_id":"1908.07290","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper proves a five-term linear independence theorem for certain Lambert series, implying the Fibonacci and Lucas reciprocal sums over prime squares are linearly independent with 1 over Q(√5).","lead":"This paper proves that 1 and the sums of reciprocals of Fibonacci and Lucas numbers over sparse coprime index sets, such as squares of primes, are linearly independent over Q(√5). It upgrades earlier irrationality results to a stronger independence statement using a refined version of Erdos's congruence method.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proof of Theorem 1.1 has a gap: the leading coefficient D_k need not tend to infinity, so the lower bound on |Θ_k| is not justified.","rationale":"The reader’s ACK with moderate confidence focuses on the explicit coprimality hypothesis (H1) as the weakest assumption. I agree that (H1) is critical and that removing it would break the construction, but it is an explicit hypothesis of the theorem, not a hidden gap. My stress-test instead identifies a gap in the internal proof: the claim that |D_k| → ∞ whenever the ρ_j are not all zero is not generally true, and the case where D_k is constant or identically zero is not addressed. This affects the central contradiction at (3.6), which needs a nonzero Θ_k for the norm argument. The gap is repairable if one can show that for any nontrivial relation the coefficient D_k is nonzero (or if one modifies the construction to avoid the degenerate case), but the paper as written does not supply such an argument. Since the main theorem’s proof is incomplete at this point, I would not accept the paper as-is; I would require the authors to justify the nonvanishing of the leading term or add a treatment of the degenerate case. The published result is plausible and the rest of the construction appears sound, which is why I recommend CONDITIONAL rather than REJECT.","tokens_in":13763,"tokens_out":40383,"duration_ms":365801,"concrete_test":"Fix α = (1+√5)/2 and take the nonzero algebraic integers ρ1 = 2α^2, ρ2 = -α^3, ρ3 = ρ4 = 0. Compute the coefficient A = α^{-4}(ρ1 α^3 + 2ρ2 α^2). This equals 0, so D_k is constant. Verify that the constant term B is nonzero by direct substitution into (3.1); this disproves the claim that |D_k| → ∞ for all non-zero ρ. Then check whether one can also impose B = 0 (for example by additionally choosing ρ3, ρ4 appropriately) to obtain D_k ≡ 0 with ρ not all zero. If such ρ exists, the proof's lower-bound argument in (3.6) fails, and the contradiction must be repaired by a different mechanism.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"In the proof of Theorem 1.1, after equation (3.1), the authors define D_k as the coefficient of (2/α^8)^{k-1} in Θ_k and state: “Note that |D_k| → ∞ (k → ∞), since the ρ_j are not all zero.” This statement is false as written. From the definition, D_k = k A + B with A = α^{-4} ∑_{j=1}^4 j ρ_j α^{4-j}. If A = 0 and B ≠ 0, then D_k is a nonzero constant and |D_k| does not tend to infinity; if A = B = 0, then D_k ≡ 0. The subsequent lower bound 0 < |Θ_k| < k^5 (2/|α|^8)^k relies on |D_k| being bounded away from zero or growing, and the case D_k ≡ 0 is not handled. If D_k ≡ 0 for a nontrivial linear relation, then Θ_k = O((2/|α|^8)^k), and Θ_k could vanish for all large k, making the norm N_k in (3.3) equal to zero and destroying the contradiction. No argument is given to exclude A = B = 0 for a non-zero (ρ_j). This is a genuine gap in the proof of Theorem 1.1 as stated, independent of the (H1) coprimality assumption.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves a linear-independence statement: if α is an algebraic integer with |α|>1 and all conjugates except possibly α and its complex conjugate lie in the open unit disk, then 1, f1(α^{-1}), f2(α^{-1}), f3(α^{-1}), f4(α^{-1}) are linearly independent over Q(α), where the fj are the Lambert-series generating functions in (1.2). The proof combines an Erdős-type simultaneous-congruence construction that forces prescribed divisor-count values on long intervals with a Pisot-number norm argument. Applications include linear independence over Q(√5) of 1, Σ_p 1/F_{p^2}, and Σ_p 1/L_{p^2}, and more generally of reciprocal sums over Fibonacci and Lucas numbers indexed by pairwise coprime sequences.","tokens_in":14042,"tokens_out":16974,"duration_ms":175292,"significance":"If the proof is correct, the result is a substantial generalization of Erdős’s irrationality theorem and gives new linear-independence results for Lambert series and reciprocal sums of Fibonacci and Lucas numbers. The block construction in Lemmas 2.1–2.3 is intricate and largely self-contained, and the use of the Pisot condition via the Siegel–Chamfy lower bound is elegant. The paper is not accompanied by machine-checked proofs, but the algebraic computations in Section 3 are explicit and checkable. The main concern is a gap in the nonvanishing argument for Θ_k, which is load-bearing for the norm contradiction.","major_comments":[{"comment":"The assertion after (3.1) that |D_k| → ∞ because the ρj are not all zero is false as written. From (3.1) one has D_k = kA+B with A = α^{-4} Σ_{j=1}^4 jρjα^{4-j} and B = Σ_{j=1}^4 ρj(α^{j-4-2εj} - (α^2+1)(α^4+1)α^{1-εj}/(α^8-2)). If A=0 and B≠0, then D_k is a nonzero constant and does not tend to infinity; if A=B=0, then D_k≡0. The proof gives no reason why a nontrivial relation (ρj) satisfying the assumed dependence cannot lie in this kernel, and the system A=B=0 is a homogeneous linear system of two equations in four unknowns over Q(α) with nonzero solutions. This affects the crucial lower bound 0<|Θ_k| in (3.6): if D_k=0 then Θ_k is only O((2/|α|^8)^k), and the argument that Θ_k does not vanish collapses. Even in the constant case A=0, B≠0, the claimed growth is not available, and no separate argument is supplied to prove Θ_k≠0 for all large k. The norm contradiction in (3.3) requires N_k to be a nonzero integer; without a valid nonvanishing proof for Θ_k, the contradiction is not established.","section":"§3, Eq. (3.1)–(3.6)"}],"minor_comments":[{"comment":"Proposition 3.1 refers to “E1 and E2” and “one of E1 and E2,” but the sets defined in (1.8) are E1 and E3; this is a typo that should be corrected.","section":"§3, Proposition 3.1"},{"comment":"In the displayed inequality just before (2.9), the expression “k42k−1” should presumably be “k·4·2^{k-1}” or similar; please clarify the notation.","section":"§2.1, after (2.9)"},{"comment":"The inequality B_k ≥ 128k is stated after using (2.16); the derivation would be clearer if the intermediate step A_k ≥ 1 or δ_k < 1 were explicitly noted.","section":"§2.1, Eq. (2.17)"}],"recommendation":"major_revision","confidential_remarks":"The reported gap is substantive and affects the central proof, but it may be repairable with an additional argument excluding the degenerate case D_k≡0 or proving nonvanishing of Θ_k by another method. The rest of the construction appears coherent, so I do not recommend rejection at this stage. I would ask the authors to supply a rigorous nonvanishing argument for Θ_k before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read. Theorem 1.1 is a real extension of Erdős's irrationality method to a simultaneous linear independence statement for four Lambert-type series over Q(α). The block construction in Section 2 is intricate and, as far as I can tell, sound; the Fibonacci/Lucas applications are natural and the example with p^2 is attractive. No circularity, no fitted constants.\n\nThe soft spot is at the end of the proof of Theorem 1.1. After (3.1), the authors define D_k and claim that |D_k| → ∞ since the ρ_j are not all zero. That claim is false as written. D_k = kA + B, with A = Σ j ρ_j α^{-j} (up to a harmless factor) and B a similar constant. If A = 0 but B ≠ 0, then D_k is a nonzero constant, and |D_k| doesn't go to infinity. That case is actually harmless: |Θ_k| is then about |B|(2/|α|^8)^k, still nonzero, and the bound |Θ_k| < k^5(2/|α|^8)^k holds. The real problem is A = B = 0, so D_k ≡ 0. Then Θ_k = O((2/|α|^8)^k) and the proof gives no lower bound; Θ_k could vanish for large k. The norm argument needs |Θ_k| > 0 to get a nonzero rational integer N_k. The authors give no reason why A and B can't both vanish for a nontrivial relation; in a field of degree > 1, the equations Σ j ρ_j α^{-j} = 0 and the corresponding constant equation have nonzero solutions. So the contradiction is not established.\n\nThe reader's report missed this, giving MODERATE confidence and ACCEPT. I think that's too optimistic. The result may well be true, and the gap might be fixable—for example, by showing that if the first few leading terms vanish, a different choice of the ρ_j or a limiting argument still gives a nonzero Θ_k. But as written, the proof is incomplete.\n\nBottom line: this deserves a serious referee, because the construction is substantial and the theorem is likely correct. I would not accept the current version; request a revision that either proves D_k cannot be identically zero or handles that case separately. I'd bring it to a reading group to discuss the gap, but I wouldn't cite it in its present form.","headline":"New linear independence theorem with a clever Erdős-type construction, but the proof has a genuine gap at the end: the assertion |D_k| → ∞ is false, so the contradiction isn't established.","tokens_in":14549,"tokens_out":6709,"would_cite":false,"duration_ms":65114,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11J72","11A41"],"pacs":[],"model":"deepseek-v4-flash","headline":"For any algebraic integer with all other conjugates inside the unit disk, five associated power-series values are linearly independent over the field they generate.","keywords":["linear independence","Lambert series","Fibonacci numbers","Lucas numbers","Pisot numbers","reciprocal sums","Chinese remainder theorem"],"falsifier":"Let $\\alpha=(1+\\sqrt{5})/2$ and take $n_\\ell = p_\\ell^2$ (squares of odd primes). Compute $S_1=\\sum_p 1/F_{p^2}$ and $S_2=\\sum_p 1/L_{p^2}$ to high precision and use lattice reduction over the ring of integers of $\\mathbb{Q}(\\sqrt{5})$ to search for a nontrivial relation $a + b S_1 + c S_2 = 0$ with small coefficients. The theorem predicts no such relation exists; a discovered relation would disprove the corollary.","tokens_in":13571,"feed_emoji":"","tokens_out":12745,"duration_ms":108964,"temperature":0.7,"pith_summary":"This paper establishes a linear independence theorem for special power series whose coefficients count divisors coming from a fixed sequence of pairwise coprime odd integers. When that sequence is the squares of the primes, these series become sums of reciprocals of Fibonacci and Lucas numbers with prime-square indices. As a concrete consequence, the three numbers $1$, $\\sum_p 1/F_{p^2}$, and $\\sum_p 1/L_{p^2}$ are linearly independent over $\\mathbb{Q}(\\sqrt{5})$. The result refines a classical irrationality method into a general linear-independence statement over number fields.","feed_headline":"Fibonacci and Lucas reciprocal sums are independent","feed_subtitle":"Sums over prime squares of 1/F and 1/L, plus 1, are linearly independent over Q(√5).","key_machinery":"The key machinery is a block-construction method. For each large parameter $k$, the proof builds a system of simultaneous congruences modulo the first many terms of $\\{n_\\ell\\}$, chosen so that the Chinese remainder theorem produces an arithmetic progression $G(k)$ of length $B_k$ such that, for each position $m$ in a block of length $8k-4$, the coefficients $b_j(\\gamma+m)$ take prescribed powers of 2 (or zero), while outside the block they are bounded by $\\xi^{|i|}$ for any fixed $\\xi>1$. Inserting these controlled values into $f_j(\\alpha^{-1})$ isolates a dominant term proportional to $(2/\\alpha^8)^k$; a norm argument over $\\mathbb{Q}(\\alpha)$ then forces a contradiction if the five numbers were linearly dependent. The construction uses inclusion-exclusion and the prime number theorem to show that most elements of $G(k)$ avoid divisibility by the large terms of the sequence.","core_discovery":"The central discovery is Theorem 1.1: if $\\alpha$ is an algebraic integer with $|\\alpha|>1$ whose conjugates over $\\mathbb{Q}$ other than itself and its complex conjugate lie in the open unit disk, then the five numbers $1, f_1(\\alpha^{-1}), f_2(\\alpha^{-1}), f_3(\\alpha^{-1}), f_4(\\alpha^{-1})$ are linearly independent over $\\mathbb{Q}(\\alpha)$. The functions $f_j(z)=\\sum_{n\\geq 1} b_j(n)z^n$ are defined through integer coefficients $b_j(n)$ that count, among the divisors of $n$ belonging to a fixed sequence $\\{n_\\ell\\}$ of odd pairwise coprime integers, those in prescribed residue classes modulo 4. In the special case $\\alpha=(1+\\sqrt{5})/2$ and $\\{n_\\ell\\}$ the squares of odd primes, the combinations $f_1\\pm f_3$ (with appropriate normalizations) reduce to the sums $\\sum_p 1/F_{p^2}$ and $\\sum_p 1/L_{p^2}$, giving the abstract's example. The proof also yields Corollaries 1.1 and 1.2, which generalize the earlier irrationality theorem to linear independence of four Lambert-type series over $\\mathbb{Q}$ and of three reciprocal Lucas-sequence sums over $\\mathbb{Q}(\\alpha)$.","pith_inferences":["The block-construction argument is likely to extend to sequences defined by any nondegenerate Lucas sequence of the first and second kind, since only the Binet form is used in relating the Lambert series to $f_j$.","One could replace the four residue classes modulo 4 by more classes, obtaining linear independence of more than five numbers at the cost of longer blocks and sharper estimates.","The norm argument's rate of decay suggests a quantitative lower bound for linear forms in these numbers (an independence measure) is obtainable by the same method.","A natural testable extension is to algebraic integers whose conjugates lie in a prescribed region beyond the unit disk; the method's estimates would need adjusting."],"forward_implications":["For every integer $m \\geq 2$, the three numbers $1$, $\\sum_p 1/F_{p^m}$, and $\\sum_p 1/L_{p^m}$ are linearly independent over $\\mathbb{Q}(\\sqrt{5})$.","For any integer $t$ with $|t|>1$, the four numbers $1$, $\\sum 1/(t^{n_\\ell}-1)$, $\\sum 1/(t^{n_\\ell}+1)$, $\\sum t^{n_\\ell}/(t^{2n_\\ell}-1)$ are linearly independent over $\\mathbb{Q}$ for every sequence satisfying (H1) and (H2).","The theorem applies to all Pisot numbers and complex Pisot numbers, giving linear independence over their respective fields.","The proof provides a new general method for proving linear independence of values of Lambert-type series, beyond the particular Fibonacci/Lucas setting."],"supporting_citations":[{"why":"Supplies the original block-construction method for proving irrationality of certain Lambert series, which this paper refines into a linear-independence proof.","marker":"[7]"},{"why":"Establishes that the smallest Pisot number is about 1.3247, used to guarantee $|\\alpha|^5 > 2$.","marker":"[12]"},{"why":"Establishes the smallest modulus of a complex Pisot number is $\\sqrt{\\theta_0}$, giving the same lower bound in the non-real case.","marker":"[4]"}],"fun_headline_variants":["Prime-square reciprocal sums prove independent","Fibonacci-Lucas reciprocal sums independent over Q(√5)","Independence of Fibonacci and Lucas reciprocal sums over prime squares","Prime-squared Fibonacci and Lucas reciprocal sums independent"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The sequence of integers $\\{n_\\ell\\}$ must be pairwise coprime (condition (H1)); without this, the Chinese remainder theorem step and the inclusion-exclusion count of 'bad' positions in the block construction both fail.","fun_headline_variants_meta":{"raw":{"variants":["Prime-square reciprocal sums prove independent","Fibonacci-Lucas reciprocal sums independent over Q(√5)","Independence of Fibonacci and Lucas reciprocal sums over prime squares","Prime-squared Fibonacci and Lucas reciprocal sums independent"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001394,"raw_usage":{"total_tokens":5643,"prompt_tokens":951,"completion_tokens":4692,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":567,"completion_tokens_details":{"reasoning_tokens":4633}},"tokens_in":567,"tokens_out":4692,"duration_ms":33391,"temperature":1.0,"reasoning_tokens":4633,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:22:00.020567+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Let $\\alpha=(1+\\sqrt{5})/2$ and take $n_\\ell = p_\\ell^2$ (squares of odd primes). Compute $S_1=\\sum_p 1/F_{p^2}$ and $S_2=\\sum_p 1/L_{p^2}$ to high precision and use lattice reduction over the ring of integers of $\\mathbb{Q}(\\sqrt{5})$ to search for a nontrivial relation $a + b S_1 + c S_2 = 0$ with small coefficients. The theorem predicts no such relation exists; a discovered relation would disprove the corollary.","supporting_citations":[{"cited_title":"Erd˝ os,On the irrationality of certain series","cited_arxiv_id":null,"evidence_quote":"Supplies the original block-construction method for proving irrationality of certain Lambert series, which this paper refines into a linear-independence proof."},{"cited_title":"Siegel, Algebraic integers whose conjugates lie in the unit circle","cited_arxiv_id":null,"evidence_quote":"Establishes that the smallest Pisot number is about 1.3247, used to guarantee $|\\alpha|^5 > 2$."},{"cited_title":"Chamfy, F onctions m´ eromorphes dans le cercle-unit´ e et leurs s´ eries de Taylor","cited_arxiv_id":null,"evidence_quote":"Establishes the smallest modulus of a complex Pisot number is $\\sqrt{\\theta_0}$, giving the same lower bound in the non-real case."}],"review_version":1}