REVIEW 7 minor 8 references
Every real number is a sum of two real numbers with diverging partial quotients
T0 review · 0 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash
Pith's one-line read Every real number can be written as the sum of two numbers whose continued-fraction partial quotients go to infinity.
desk verdict The strict inequality c_n > b_n is the right lever: it turns Shulga's decomposition into a proof that both summands have diverging partial quotients, and the main theorem is sound; only auxiliary computer checks and a compressed algebra step need cleanup. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The engine of the proof is an interval-intersection reformulation of the algorithm. For candidate digit blocks, sets B_k and C_k are intervals in [0,1] that encode the condition that the first k digits produced by the algorithm equal the given blocks, and the true input α lies in their intersection A_n for every n. The proof converts the existence of a common point into explicit inequalities between continued-fraction convergents, most importantly Lemma 9(4)–(5), which under the assumption c_n = b_n yield contradictory bounds z < (1+1/y)(1-x) and z > (1+1/y)(1-x). This contradiction gives the strict inequality c_n > b_n, and Lemma 12 converts the resulting growth of t_n/q_n into divergence of b_n and c_n.
What would settle it
Run the algorithm on every rational p/q with q up to some large bound and check whether, for any n > 1, the computed digits satisfy c_n ≤ b_n; Theorem 4 predicts none exists, so any violation would refute the main inequality. Equally decisive: verify the explicit nesting construction claimed for Theorem 7 by computing its first digits [0;2,6,11,16,21,26,...] and [0;4,9,14,19,24,28,...] and checking that the stated containments B_n ⊇ C_n ⊇ B_{n+1} hold for all n; a single failed containment would break the construction.
Extended reading notes
Core claim
The central claim is Theorem 1: for every real α there exist β, γ in the set G (rationals, or irrationals whose partial quotients a_n(·) tend to infinity) with α = β + γ. For α in [0,1] the decomposition is produced by the algorithm studied here: it defines digit sequences b_n and c_n and stops early if α already equals the sum of the finite continued fractions; otherwise β = [0;b_1,b_2,...] and γ = [0;c_1,c_2,...]. The paper's main technical discovery is that the second sequence strictly dominates the first, c_n > b_n for every n > 1, which forces the ratios t_n/q_n of the convergent denominators to grow without bound; from this the divergence of the partial quotients follows. For rational α, the same denominator growth forces the algorithm to terminate, giving rational β and γ.
Load-bearing premise
The argument that the algorithm stops for rational inputs assumes a standard denominator bound: a rational number lying in a continued-fraction cylinder interval has reduced denominator at least the denominator of the interval's endpoint, and the proof uses this without stating it explicitly.
Editorial extensions
If this is right
- If the theorem is correct, the set G of rationals and numbers with diverging partial quotients is additively universal: G + G = R.
- For rational α = p/q the algorithm terminates, and the continued-fraction lengths of β and γ are O(q^2); the paper notes numerical evidence that the true length may be O(log q).
- The strict inequality c_n > b_n forces b_n ≥ n for every n, so the constructed numbers have explicit, at least linear growth in their partial quotients.
- The constructed example satisfying 4n − 2 ≤ b_n < c_n < 5n shows that linear growth is best possible up to a constant factor and that the c_n sequence need not be monotone, settling an open problem.
Reading between the lines
- The same interval-inequality machinery may transfer to other continued-fraction classes, such as numbers whose partial quotients stay above a prescribed increasing function; the strict c_n > b_n inequality is likely the flexible heart of the argument.
- One could test computationally whether the O(log q) bound for rational inputs truly holds; the paper leaves this open, and a proof or counterexample would sharpen the constructive claim.
- Since G has Hausdorff dimension 1/2, the theorem shows that a small-dimensional set of continued-fraction numbers can still have a sumset covering the whole real line, in contrast with the bounded-partial-quotient sums in Hall's classical result.
- The explicit nesting construction in Theorem 7 could be adapted to produce decompositions with prescribed linear growth rates, giving a family of examples rather than a single one.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proves that every real number can be written as a sum of two numbers whose continued-fraction partial quotients either terminate (for rationals) or tend to infinity (for irrationals). The proof is constructive and is based on Shulga's decomposition algorithm. The main technical contribution is Theorem 4, which strengthens Shulga's inequality c_n >= b_n to the strict inequality c_n > b_n for n > 1. From this the authors derive that t_n/q_n tends to infinity (Lemma 12), which yields Corollary 5: for every alpha in [0,1], the Shulga decomposition beta + gamma has beta, gamma in G, and if alpha is rational then beta and gamma are rational. The paper also proves growth estimates for the partial quotients (Theorem 6) and constructs an irrational example with linear growth (Theorem 7), giving a negative answer to Shulga's Problem 6.1. The proof of the main theorem is self-contained: Section 3 re-proves Shulga's theorem using interval intersections rather than assuming it.
Significance. If correct, the paper resolves a natural question posed by Shulga and adds to the classical line of results by Hall, Erdős, and Cusick on sums of sets defined by continued-fraction constraints. The proof is constructive and the interval-intersection analysis of the Shulga algorithm is of independent interest. The main argument is largely self-contained and derives the key inequalities in detail. The main theorem is not affected by the auxiliary gaps I note below, which concern Theorem 6 and the wording of Lemma 14. I found no circularity: the paper re-proves Shulga's theorem and does not assume the target result.
minor comments (7)
- [§5, proof of Theorem 6] The assertion that "one can easily check (e.g., by a computer program)" the base cases n <= 6, in particular b6 >= 8, is not supported by any reproducible verification in the manuscript. Since the induction for (5) and (6) uses (34) as its base, this is an omitted proof for a stated theorem. The same applies to the "direct computation" for 2 <= n <= 6 in Lemma 15. This gap does not affect the proof of Theorem 1, but it should be repaired by supplying an explicit table, a short human proof, or well-documented code.
- [§5, Lemma 14] The sentence "which is equivalent to c_{n+1}/(b_n - 1) > 1 + c_{n+1}/t" before equation (33) is inaccurate: the displayed algebra gives (33) as a one-way consequence of the preceding inequality, not as an equivalent reformulation. The later deductions use only the one-way implication, so the proof remains valid after rewording, but the current text is formally incorrect.
- [§4, proof of Corollary 5] The line "from (3) we have q_{n+1}q_n > t_n^2" is not a direct consequence of the inequality c_n > b_n. It follows from Lemma 9(3) applied to the nonempty intersection B_{n+1} ∩ C_n for alpha in A_{n+1}, which gives q_{n+1}q_n > t_n(t_n + t_{n-1}) > t_n^2. Please correct the citation or add the missing step.
- [§4, proof of Corollary 5, rational case] The proof uses the standard but unstated fact that a rational number lying in the cylinder interval [[0;c_1,...,c_n], [0;c_1,...,c_n+1]] has reduced denominator at least t_n. This fact is correct, but it should be stated explicitly, because the bound t_n/q_n <= q depends on it.
- [§3, Lemma 12] The chain "From (15) one can see that c_n/(2t_n^2) < 1/(2t_{n-1}t_n) < 2/q_n^2" is compressed. The second inequality uses q_n - q_{n-1} >= q_n/2, which follows from b_n >= 2. This intermediate step should be written out for readability.
- [§1, Corollary 5 implies Theorem 1] The implication from Corollary 5 to Theorem 1 is not explained in the text. It should be noted explicitly that adding an integer to a real number in G preserves membership in G, so the result for [0,1] extends to all of R by an integer shift.
- [Various] There are several typos and minor formatting issues: "Defition 2" in Section 4; "Let n > 6 be an arbitrary integer greater than 6" in the proof of Theorem 6; the displayed expression "4 − 3.8(c_n−x)/(c_n+2.1) (c_n−x)+x" in Lemma 15 is garbled and should read 4 + x − 3.8(c_n−x)/(c_n+2.1); and the reference to "Erd˝ os" should use the standard accent.
Circularity Check
No significant circularity: the main theorem is proved by a self-contained analysis of Shulga's algorithm, with no fitted parameters and no prediction that reduces to an input.
full rationale
The paper's central claim (Theorem 1) is derived from Corollary 5, which is derived from Theorem 4 and Lemma 12. These results are proved from Lemmas 8-11, and Lemma 10 explicitly re-proves Shulga's Theorem 3 in a self-contained way: the proof uses only the interval containment argument and Lemma 9, not the cited theorem as an assumption. In particular, the inequality c_k >= b_k used inside Lemma 10 is obtained from Lemma 9(4), whose hypotheses follow from alpha in A_n; it is not imported from [8]. Theorem 4's proof is by contradiction using Lemmas 9(4)-(5), both proved in the paper. The rational-termination step in Corollary 5 uses a standard continued-fraction denominator fact (a rational in a cylinder has reduced denominator at least the cylinder denominator), which is an auxiliary background fact, not a restatement of the conclusion. There are no fitted parameters, no subset of data predicted from a fit, and no uniqueness theorem invoked from the authors' prior work. The only self-citations are to Shulga's algorithm and problems, which are used as motivation; the load-bearing part is re-proved. The asserted computer checks for n <= 6 in Theorem 6 are unverified in the text but are finite base cases, not circularity. Thus the derivation chain is self-contained and no circular step can be exhibited.
Assumptions & free parameters
assumptions (3)
- standard math Standard continued fraction cylinder and convergence identities, equations (11) and (12).
- standard math Continued fraction convergent recurrences and determinant identity, equations (8) through (10).
- domain assumption A rational number in a cylinder interval with endpoint denominator t_n has reduced denominator at least t_n.
Cite this review
Pith. "Pith review of Every real number is a sum of two real numbers with diverging partial quotients." pith.science (2026). https://pith.science/paper/XRX3B5NU
@misc{pith2026250704521,
author = {Pith},
title = {Pith review of: Every real number is a sum of two real numbers with diverging partial quotients},
year = {2026},
howpublished = {\url{https://pith.science/paper/XRX3B5NU}},
note = {Machine review of arXiv:2507.04521}
}
read the original abstract
We show that every irrational number is a sum of two real numbers with diverging partial quotients. The proof is constructive. The key towards these results is an algorithm which was recently developed by Nikita Shulga, and our study of this algorithm is of independent interest.
Figures
Reference graph
Works this paper leans on
-
[1]
Thomas W. Cusick. Sums and products of continued fractions.Proceedings of the American Mathematical Society, 27(1):35–38, 1971
work page 1971
-
[2]
Representations of real numbers as sums and products of Liouville numbers
Paul Erd˝ os. Representations of real numbers as sums and products of Liouville numbers. Michigan Mathematical Journal, 9(1):59 – 60, 1962
work page 1962
-
[3]
Irving J. Good. The fractional dimensional theory of continued fractions.Proc. Cambridge Philos. Soc., 37:199–228, 1941
work page 1941
-
[4]
On the sum and product of continued fractions.Annals of Mathematics, 48(4):966–993, 1947
Marshall Hall. On the sum and product of continued fractions.Annals of Mathematics, 48(4):966–993, 1947
work page 1947
-
[5]
Alexander.Y.Khinchin. Continued fractions. DoverPublications,Inc.,Mineola,NY,russian edition, 1997. With a preface by B. V. Gnedenko, Reprint of the 1964 translation
work page 1997
-
[6]
Rockett and Peter Sz¨ usz.Continued fractions
Andrew M. Rockett and Peter Sz¨ usz.Continued fractions. World Scientific Publishing Co., Inc., River Edge, NJ, 1992
work page 1992
-
[7]
Schmidt.Diophantine approximation, volume 785 ofLecture Notes in Mathe- matics
Wolfgang M. Schmidt.Diophantine approximation, volume 785 ofLecture Notes in Mathe- matics. Springer, Berlin, 1980
work page 1980
-
[8]
On a Conjecture of Cusick on a sum of Cantor sets
Nikita Shulga. On a conjecture of Cusick on a sum of Cantor sets, 2024. arXiv:2411.17379. 18
work page Pith review arXiv 2024
Reviewed August 6, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.