{"id":"e3e1935b-7c3d-4922-aec6-85dab5001b07","arxiv_id":"2411.17379","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Cusick's conjecture is false: S(k)+S(k) always contains the interval [0,1/(k-1)].","lead":"This paper disproves a fifty-year-old conjecture by Cusick: for every k at least 2, the set of numbers expressible as a sum of two continued fractions with all partial quotients at least k contains the whole interval from 0 to 1/(k-1). The proof is constructive and also identifies infinitely many gaps in the sumset, giving a fuller picture of how these Cantor-like sets add.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the compressed even-parity arguments in Lemmas 3.4, 3.6, 3.7, 3.8 are symmetric and check out.","rationale":"The reader's weakest assumption correctly identifies the compressed parity cases as the most delicate part of the proof. I therefore examined those cases directly. The even-index version of Lemma 3.4 follows from the same distance estimates as the odd case, with the interval inclusions for I_{2m} and I_{2m-1} swapped; the resulting bound is exactly (16). The even case of Lemma 3.6 reduces to an inequality between denominators that is a consequence of b_i ≥ c_i, which itself is established by induction using Lemma 3.8 on earlier indices. I also checked the logical order of Lemmas 3.6 and 3.8: Lemma 3.8 for index k-1 gives b_i ≥ c_i for i ≤ k without needing c_{k+1}, so the proof is not circular. The central claim, Theorem 1.2, is supported by the algorithm's correctness, convergence, and lower partial-quotient bounds derived in the lemmas. The remaining issues are stylistic: odd/even cases are not fully written out, and a few typos appear. None of these undermine the conclusion. Hence I find no load-bearing objection, and the reader's ACCEPT verdict stands unchanged.","tokens_in":17157,"tokens_out":54682,"duration_ms":365890,"concrete_test":"Independently write out the even-index case of Lemma 3.4 using Proposition 2.1, and verify that the final inequality is q_{2m}q_{2m-1} > t_{2m-1}(t_{2m-1}+t_{2m-2}); if this matches (16), the parity concern is settled.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I attempted to locate a failure point in the algorithmic construction behind Theorem 1.2. The least-secure spot is the paper's repeated reliance on 'the same argument with inequality signs reversed' for even indices in Lemmas 3.4, 3.6, and 3.7, and for the even-n case of Theorem 1.3. Working out the even case of Lemma 3.4: with n=2m, the definition of c_{2m} gives (p_{2m}-p_{2m-1})/(q_{2m}-q_{2m-1}) ≤ x - s_{2m-1}/t_{2m-1} < p_{2m}/q_{2m}, and the definition of b_{2m-1} puts y = x - p_{2m-1}/q_{2m-1} in ((s_{2m-1}+s_{2m-2})/(t_{2m-1}+t_{2m-2}), s_{2m-1}/t_{2m-1}]. Combining the implied distance bounds gives 1/(q_{2m}q_{2m-1}) < 1/(t_{2m-1}(t_{2m-1}+t_{2m-2})), exactly (16). The even case of Lemma 3.6 similarly reduces to Q_{2n+1}(Q_{2n+1}+Q_{2n}) < t_{2n+1}(t_{2n+1}+t_{2n}), which follows from b_i ≥ c_i and strict denominator growth. The apparent circularity between Lemmas 3.6 and 3.8 is resolved by induction: before proving c_{k+1}, apply Lemma 3.8 for indices < k to obtain b_i ≥ c_i for i ≤ k; then finish Lemma 3.6. I found no sign error, no hidden assumption, and no fitted parameter. The main theorem is supported by the lemmas as written, apart from expositional compression and minor typos (e.g., '2k' in Lemma 3.4 should read '2m'; a missing minus sign in the final numerical display).","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves that Cusick's conjecture is false by showing that for every k≥2 the sumset S(k)+S(k) contains the interval [0,1/(k−1)], so its Lebesgue measure is at least 1/(k−1). The proof is constructive: an iterative algorithm, defined from the continued fraction digits of x and of the residual x−s_n/t_n, produces two numbers whose partial quotients are all at least k and whose sum converges to x. The paper also constructs countably many gaps in S(k)+S(k), proves containment results for S(m)+S(n), and gives an exact interval statement S(m)+S(m^2)=[0,(m+1)/m^2].","tokens_in":17566,"tokens_out":15931,"duration_ms":128661,"significance":"If the proof is correct, this is a significant resolution of a long-standing conjecture: the sum of two copies of S(k) has positive measure for every k≥3, contrary to Conjecture 1.1. The proof is self-contained and algorithmic, and it does not depend on any numerical fitting or on assuming the conjecture. The lower-bound chain b_n≥c_n and c_{n+1}≥(c_1−1)^2 is the engine of the argument, and the apparent circularity between Lemmas 3.6 and 3.8 is resolved by an induction on the index. The paper also recovers Cusick's S(2)+S(2)=[0,1] as a corollary, which is a useful consistency check.","major_comments":[],"minor_comments":[{"comment":"The even-index cases are not written out: Lemma 3.4 says that for even indices the process is exactly the same except for inequality signs, and Lemma 3.7 says the same. Since these inequalities are used for every index in the induction, please include the even-index verification or at least a precise statement of which inequalities reverse. I checked the even case of Lemma 3.4 and it does reduce to (16), so this is a completeness issue rather than a mathematical error.","section":"Section 3, Lemmas 3.4 and 3.7"},{"comment":"The proof of the even-n case is deferred with the sentence \"The case of even n can be done similarly with minor changes.\" Because Theorem 1.3 is one of the main results, the even-n argument should be written out or explicitly reduced to the odd-n case by a clear symmetry.","section":"Section 5, proof of Theorem 1.3"},{"comment":"The statement says that inequality (16) holds for all 1≤n≤k, but for n=1 the right-hand side contains t_{n−2}=t_{−1}, which is undefined; the intended range appears to be 2≤n≤k. In the same proof, the phrase \"all partial quotients b_i with indices smaller or equal to 2k\" should read \"2m\".","section":"Lemma 3.4"},{"comment":"In the third numerical example, the displayed error should be |2(√2−1) − [2,51,139299,23380586] − [3,2143,8527219,38512412]|; a minus sign is missing before the second continued fraction.","section":"Section 6, numerical data"},{"comment":"The line \"by (25) for k=1 we get c_2 > 4 + 1/2\" does not follow directly from (25), which with c_1=2 and b_1≥4 gives c_2 > b_1^2/4; the displayed constant appears to be a typo, although the conclusion c_2≥5 remains valid.","section":"Proof of Theorem 1.6"},{"comment":"The symbol n is used both for the parameter in S(n) and as the running index in expressions such as c_{n+1} ≥ (c_1−1)^2; this should be disambiguated, for example by writing c_{r+1} for the running index.","section":"Proof of Theorem 1.5"},{"comment":"The initialization of the sequences uses q_0, t_0, s_0, and p_0 without an explicit definition; please state the standard conventions p_0=0, q_0=1, s_0=0, t_0=1 before the first use.","section":"Section 3"}],"recommendation":"minor_revision","confidential_remarks":"The paper is within the scope of the journal and the attribution to Cusick's conjecture is correct. The main theorem appears sound; the required changes are local. I would be happy to see the revised version without further external review."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this paper actually kills Cusick's 1971 conjecture. The main result is that for every k ≥ 2, S(k)+S(k) contains the interval [0,1/(k-1)], so its Lebesgue measure is at least 1/(k-1), directly contradicting the conjecture that it is zero for k ≥ 3. The proof is constructive: it gives an iterative algorithm that decomposes any x in that interval into two continued fractions with all partial quotients ≥ k. I worked through the main line and it holds. The algorithm is genuinely new and self-contained; it recovers Cusick's S(2)+S(2)=[0,1] as a corollary, and it also yields exact results like S(m)+S(m^2)=[0,(m+1)/m^2] and a theorem about gaps in S(k)+S(k) that rules out the natural \"maximal interval\" alternative. There are no fitted parameters and no circularity: the lower-bound chain bn ≥ cn and cn+1 ≥ (c1−1)^2 follows from explicit estimates on convergents and continuants, not from assuming the conjecture.\n\nThe soft spots are mostly expositional. The even-index cases in Lemmas 3.4, 3.6, and 3.7, and the even-n case of Theorem 1.3, are sketched with phrases like \"the same argument with inequality signs reversed.\" I checked one even case and it works; the symmetry is real, but the paper would be much easier to verify if those cases were written out. There are also typos: the '2k' in the proof of Lemma 3.4 should be '2m', and the numerical example for sqrt(2) has a missing minus sign in the error display. The \"easy to check\" inequality near the end of Lemma 3.10 is left to the reader; it is elementary but should be shown. None of these issues threatens the central argument. The proof is not machine-checked, so moderate confidence is appropriate, but the construction is explicit enough that an independent verification is feasible.\n\nThis paper is for specialists in continued fractions, arithmetic sums of Cantor sets, and fractal geometry. It deserves a serious referee: the result settles a half-century-old conjecture, the method is new, and the proof is concrete. I would send it to peer review, and I would cite it if I worked in this area. My only substantive request to the authors would be to expand the compressed parity arguments and fix the typos.","headline":"Cusick's conjecture is false: Shulga's constructive algorithm shows S(k)+S(k) contains [0,1/(k-1)], and the proof checks out despite some compressed parity cases.","tokens_in":18157,"tokens_out":2445,"would_cite":true,"duration_ms":21649,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11A55","11J70","28A80"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that for every integer $k \\geq 2$, the sumset $S(k)+S(k)$ contains the interval $[0, 1/(k-1)]$, which gives Lebesgue measure at least $1/(k-1)$ and disproves Cusick's conjecture that this measure is zero for every $k \\geq…","keywords":["continued fractions","Cantor sets","arithmetic sums of sets","Cusick's conjecture","partial quotients","Lebesgue measure","gaps in sumset"],"falsifier":"Implement the algorithm from equations (2)–(3) for $k=3$ on a fine grid of points in $(0, 1/2)$, say $x=1/3$, and check whether every produced partial quotient is at least $3$ and whether the residuals stay inside the predicted cylinders; the first failure, or a computed $c_n$ or $b_n$ below $3$, would refute Theorem 1.2. A more targeted check is to verify the even-index case of inequality (16) numerically, for example $n=2$ with $c_1=3$, $b_1=6$, $c_2=5$, since the paper only spells out the odd case.","tokens_in":16907,"feed_emoji":"➕","tokens_out":9087,"duration_ms":74456,"temperature":0.7,"pith_summary":"Cusick conjectured in 1971 that the set of reals in $[0,1]$ expressible as a sum of two continued fractions whose partial quotients avoid $1,\\dots,k-1$ has Lebesgue measure zero for every $k \\geq 3$. This paper proves the opposite: for every $k \\geq 2$ the sumset $S(k)+S(k)$ contains the whole interval $[0, 1/(k-1)]$, so its measure is at least $1/(k-1)$ and Conjecture 1.1 is false. The proof is constructive: it gives an algorithm that, for any $x$ in that interval, produces two continued fractions with all partial quotients at least $k$ and whose sum converges to $x$. The same machinery yields positive-measure results for sums $S(m)+S(n)$ of two different sets, an exact interval in the case $S(m)+S(m^2)$, and a countable family of genuine gaps in $S(k)+S(k)$. A sympathetic reader should care because the result overturns a fifty-year-old expectation and replaces a dimension-only lower bound with an explicit interval containment.","feed_headline":"Cusick's zero-measure conjecture fails for every k ≥ 3","feed_subtitle":"S(k)+S(k) contains an interval of length 1/(k−1), so its Lebesgue measure is at least 1/(k−1).","key_machinery":"The engine is an explicit iterative algorithm. Starting with $c_1 = a_1(x)+1$, it defines $b_n = a_n(x - p_n/q_n)$ and $c_{n+1} = a_{n+1}(x - s_n/t_n)+1$, where $p_n/q_n$ and $s_n/t_n$ are the convergents of the two fractions being built; the construction alternates between the two summands so that each is a good approximation to $x$ minus the other. The proof that the algorithm is well-defined, converges, and produces partial quotients $\\geq k$ reduces to a system of inequalities comparing the continuant denominators $q_n$ and $t_n$ (Lemmas 3.3–3.8), with the key amplification estimate (Lemma 3.10) showing that once $c_1 \\geq 3$, all later $c_{n+1} \\geq (c_1-1)^2$. This denominator comparison is what carries the argument; it replaces the thickness and critical-exponent machinery of earlier work with a direct construction.","core_discovery":"The central discovery is that the addition of two copies of the continued-fraction Cantor set $S(k)$ is much larger than previously believed: for any $k \\geq 2$, $S(k)+S(k)$ contains $[0, 1/(k-1)]$. Since $S(k)+S(k)$ is trivially contained in $[0, 2/k]$, this does not reach the whole maximal interval, but it gives Lebesgue measure at least $1/(k-1)$, which for $k \\geq 3$ is positive, contradicting Cusick's Conjecture 1.1. In addition, the paper shows that $S(k)+S(k)$ is not the full maximal interval by exhibiting countably many open gaps $G_{k,n}$ whose endpoints lie in the sumset but whose interiors do not, and it extends the interval-containment result to unequal sets $S(m)+S(n)$ under the condition $n \\leq (m-1)^2$, with an exact equality $S(m)+S(m^2) = [0, (m+1)/m^2]$.","pith_inferences":["Because the algorithm is explicit and converges quickly in the paper's worked examples, it may serve as a practical tool for representing real numbers as sums of two restricted continued fractions, with possible use in Diophantine approximation or digit-expansion problems where simultaneous restrictions are wanted.","The amplification estimate $c_{n+1} \\geq (c_1-1)^2$ hints that the construction is quite robust: once the first partial quotient exceeds $k$, later ones grow at least quadratically, so the interval $[0, 1/(k-1)]$ might be far from the largest interval contained in $S(k)+S(k)$, a question the paper does not attempt to optimize.","The gap pattern around $2/S_k$, twice the reciprocal metallic mean, suggests that $S(k)+S(k)$ has a periodic, self-similar boundary; characterizing the full complement inside $[0, 2/k]$ is an open direction the paper only begins.","If, as the numerical data suggest, the merged partial-quotient sequence $c_1, b_1, c_2, b_2, \\dots$ is non-decreasing (open Problem 6.1), then every real number in $(0,1]$ would be a sum of two elements from Good's set of numbers whose partial quotients tend to infinity, resolving the paper's Problem 6.2."],"forward_implications":["For every $k \\geq 2$, the Lebesgue measure of $S(k)+S(k)$ is at least $1/(k-1)$; in particular $S(3)+S(3)$ has measure at least $1/2$, so the zero-measure conjecture fails already at $k=3$.","By monotonicity of the nested sets $S(k)$, the containment $S(k)+S(k) \\supseteq [0, 1/(k-1)]$ gives the lower bound $\\lambda(S(m)+S(n)) \\geq 1/(\\max(m,n)-1)$ for any $m,n \\geq 2$ (Corollary 1.4).","Under the condition $3 \\leq m < n \\leq (m-1)^2$, the same algorithm shows $S(m)+S(n) \\supseteq [0, 1/(m-1)]$, and for the special pair $(m, m^2)$ the sum is exactly the interval $[0, (m+1)/m^2]$.","The gap theorem shows that $S(k)+S(k) \\neq [0, 2/k]$ for every $k \\geq 3$: the countably many open intervals $G_{k,n}$ lie inside the maximal interval but are disjoint from the sumset, with endpoints in the sumset."],"supporting_citations":[{"why":"States Conjecture 1.1, the zero-measure claim this paper disproves, and the $k=2$ identity $S(2)+S(2)=[0,1]$ that Theorem 1.2 generalizes.","marker":"[5]"},{"why":"Provides the cylinder endpoints and continuant identities (Proposition 2.1 and Lemma 2.2) used throughout the induction and denominator estimates.","marker":"[15]"},{"why":"Supplies the previous best lower bound on $\\dim_H(S(k)+S(k))$ and the failure of its interval criterion that motivated the conjecture; the new interval containment supersedes it.","marker":"[2]"}],"fun_headline_variants":["Cusick's Cantor-sum conjecture disproved for every k ≥ 3","Sumset S(k)+S(k) contains [0,1/(k-1)], killing Cusick's conjecture","Half-century-old zero-measure conjecture on continued fractions is false","Constructive proof shows Cusick's 1971 conjecture fails completely"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The construction works only if the denominator estimates that control the algorithm's induction hold for even indices exactly as claimed for odd indices; the paper writes out the odd-index cases of Lemmas 3.4 and 3.7 in detail and asserts the even-index cases follow by the same argument with signs reversed, so a hidden sign error there would break the decomposition for some $x$.","fun_headline_variants_meta":{"raw":{"variants":["Cusick's Cantor-sum conjecture disproved for every k ≥ 3","Sumset S(k)+S(k) contains [0,1/(k-1)], killing Cusick's conjecture","Half-century-old zero-measure conjecture on continued fractions is false","Constructive proof shows Cusick's 1971 conjecture fails completely"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000382,"raw_usage":{"total_tokens":2108,"prompt_tokens":1112,"completion_tokens":996,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":728,"completion_tokens_details":{"reasoning_tokens":907}},"tokens_in":728,"tokens_out":996,"duration_ms":8553,"temperature":1.0,"reasoning_tokens":907,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T12:12:39.659793+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Implement the algorithm from equations (2)–(3) for $k=3$ on a fine grid of points in $(0, 1/2)$, say $x=1/3$, and check whether every produced partial quotient is at least $3$ and whether the residuals stay inside the predicted cylinders; the first failure, or a computed $c_n$ or $b_n$ below $3$, would refute Theorem 1.2. A more targeted check is to verify the even-index case of inequality (16) numerically, for example $n=2$ with $c_1=3$, $b_1=6$, $c_2=5$, since the paper only spells out the odd case.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"States Conjecture 1.1, the zero-measure claim this paper disproves, and the $k=2$ identity $S(2)+S(2)=[0,1]$ that Theorem 1.2 generalizes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the cylinder endpoints and continuant identities (Proposition 2.1 and Lemma 2.2) used throughout the induction and denominator estimates."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the previous best lower bound on $\\dim_H(S(k)+S(k))$ and the failure of its interval criterion that motivated the conjecture; the new interval containment supersedes it."}],"review_version":1}