On Entry II.16.12: A continued fraction of Ramanujan

Gaurav Bhatnagar; Mourad E. H. Ismail

REVIEW 5 major objections 4 minor 16 references

On Entry II.16.12: A continued fraction of Ramanujan

T0 review · 5 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Ramanujan's Entry 12 continued fraction identity is proved by two independent methods, and its natural generalization is evaluated in closed form.

desk verdict Two credible new proofs of Ramanujan's Entry 12, plus a genuinely new generalization whose proof is not yet complete as stated. read the letter →

arxiv 1908.03333 v1 pith:OO3J7HZG submitted 2019-08-09 math.CA

Gaurav Bhatnagar , Mourad E. H. Ismail This is my paper · ORCID

classification math.CA MSC 33D4530B70

keywords continuedfractionsorthogonalpolynomialsRamanujanq-seriesEntry12J-fractionsDarboux'smethodbasichypergeometricseries

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proves Ramanujan's Entry 12: an infinite continued fraction whose k-th numerator is (a−$bq^{{2k−1}}$)(b−$aq^{{2k−1}}$) equals the ratio of q-products ($a^{2}$ $q^{3}$, $b^{2}$ $q^{3}$; $q^{4}$)_∞ / ($a^{2}$ q, $b^{2}$ q; $q^{4}$)_∞ for |q|<1 and |ab|<1. It gives two independent derivations. The first, Euler's method, divides one basic hypergeometric series by another using only the q-binomial theorem, and yields modified convergence. The second, the standard orthogonal-polynomial approach, uses Darboux's asymptotic method on generating functions of the convergents and establishes ordinary convergence. In addition, the paper evaluates the natural J-fraction generalization H(x) explicitly as 2ρF(ρ)/G(ρ), a ratio of 2φ1 series, for all complex x outside (−1,1) except possibly a finite set, under the assumption that a and b are real with opposite signs.

What carries the argument

The central objects are the J-fraction H(x) from (1.2), whose k-th partial numerator is (a−$bq^{{2k−1}}$)(b−$aq^{{2k−1}}$) and whose k-th denominator is x(1−ab)+(1−ab)$q^{{2k}}$, together with its convergent numerator and denominator polynomials. These satisfy the three-term recurrence y_{k+1}(x)=((1−ab)x+(1−ab)$q^{{2k}}$)y_k(x)+ab(1−$bq^{{2k−1}}$/a)(1−$aq^{{2k−1}}$/b)y_{k−1}(x). The proofs ride on this recurrence: it shows the denominator polynomials are orthogonal with respect to a positive measure when a and b have opposite signs, and it leads to explicit generating functions for the scaled polynomials. Darboux's method is then used to read off the asymptotics of the convergents from the dominant singularities of those generating functions, and Markov's theorem converts the resulting ratio into the closed-form 2φ1 expression. For the Euler-method proof, the load-bearing identity is the elementary division formula U/V=1+(U−V)/V applied to a carefully chosen ratio of two 2φ1 series.

What would settle it

For a=0.3, b=0.4, q=0.5, compare 20-term convergents of (1.1) with the infinite product; for a=2, b=3 (both positive), q=0.5, x=2i, compare convergents of (4.3) with 2ρF(ρ)/G(ρ).

Watch

Extended reading notes

Core claim

Ramanujan recorded Entry 12 as an equality between a continued fraction and a ratio of infinite q-products, and this paper establishes that equality in full generality and with two different proof strategies. The first strategy treats the left-hand side as a ratio of two 2φ1 basic hypergeometric series and applies Euler's method—repeatedly rewriting U/V = 1+(U−V)/V—to recover the continued fraction. The second strategy attaches to the continued fraction a J-fraction H(x), whose convergent numerator and denominator polynomials satisfy a three-term recurrence; after a change of variable and scaling, the denominators are orthogonal with respect to a positive measure on a bounded interval (when a and b have opposite signs). Darboux's method extracts the asymptotic behaviour of the convergents from their generating functions, and Markov's theorem identifies the limit of the ratio of convergents as the Stieltjes transform of that measure. The result is an explicit evaluation X(x)=2ρF(ρ)/G(ρ), with ρ chosen according to the half-plane of x, which for x=1 reduces to the value needed to finish Ramanujan's original continued fraction.

Load-bearing premise

The evaluation of the generalized continued fraction in Theorem 4.1 relies on a and b being real numbers with opposite signs, so that the recurrence coefficients β_k are positive and the denominator polynomials are orthogonal with respect to a positive measure; if that sign condition fails, the stated convergence for all complex x outside (−1,1) is not proved.

Editorial extensions

If this is right

Ramanujan's Entry 12 identity (1.1) holds unconditionally for |q|<1 and |ab|<1, with ordinary (not merely modified) convergence.
The apparent anomaly of the first denominator in Ramanujan's continued fraction is explained: the 'correct' first denominator from the orthogonal-polynomial viewpoint is 2(1−ab), and Ramanujan's fraction is recovered from it by the relation 1/K − (1−ab) = 1/C.
The generalized J-fraction H(x) is evaluated in closed form for all complex x outside the interval (−1,1) except possibly a finite set, extending the family of exactly evaluated q-continued fractions.
Values for parameters outside the original range follow by symmetry: |ab|>1 by replacing a,b with reciprocals, and |q|>1 by replacing q with 1/q.
The denominator polynomials of H(x) form an orthogonal polynomial sequence with a positive measure supported on a bounded interval, and the measure has an absolutely continuous component on (−1,1) with possible discrete mass points only outside that interval.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Beyond the paper: the success of Euler's method here suggests a general recipe for other Ramanujan continued fractions whose first denominator breaks the pattern: first identify the ratio of two basic hypergeometric series that produces the fraction, then apply the division identity U/V=1+(U−V)/V repeatedly.
Beyond the paper: since the closed form X(x)=2ρF(ρ)/G(ρ) is a Stieltjes transform of the orthogonality measure, the standard inversion formula for Stieltjes transforms can be applied to obtain an explicit density for the absolutely continuous part of the measure on (−1,1).
Beyond the paper: the opposite-signs assumption on a and b is sufficient for the measure-theoretic argument, but the final formula is analytic in the parameters; it may extend to wider parameter ranges by analytic continuation whenever the continued fraction converges, which numerical experiments could test.
Beyond the paper: the observation that Ramanujan's first denominators are 'off' by a factor of 2 may point to a systematic explanation for the shape of many of his continued fractions: he likely derived them from the product side by division, which naturally produces the anomalous first term.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Justin—quick take. This paper has two real contributions: a clean Euler-method proof of Entry 12 (the last missing entry in the program from [4]), and a second proof by Darboux/orthogonal polynomials. Both are careful, the algebra is explicit, and there is no circularity: they don't assume the identity they prove. The identity itself is Ramanujan's and was previously proved by Adiga–Berndt–Bhargava–Watson and others, so the novelty is in method, not in the equation. That's fine—the method is the point.

The generalization in Theorem 4.1 is genuinely new, as far as I can tell, but its proof is not fully supported. Two issues. First, the 2φ1 series F and G require |bq/a| and |b/(aq)| to be less than 1. The paper only assumes a,b real of opposite signs with |ab|<1, which does not guarantee either bound. Without a stated analytic continuation (for instance via the reciprocal symmetry used in Remarks after Corollary 3.2), the formula isn't even defined for all allowed parameters. Second, the convergence assertion 'for all x∉(-1,1) except a finite set' is justified by citing Markov's theorem. But Markov's theorem, as cited, gives convergence outside the true interval of orthogonality, and that interval can extend beyond (-1,1) because the measure can have isolated masses outside (-1,1). So the stated convergence on the gaps between (-1,1) and those masses doesn't follow from that citation. This might be fixable by a more refined theorem, but the paper doesn't supply it. The Darboux computations are also compressed: generating functions are stated without derivation, and the most delicate parts are 'similar to [5]' and omitted.

There is a smaller, easy fix: Corollary 3.2 is stated for |ab|<1, but the proof uses Theorem 3.1/2.1 which carry the extra condition |a^2 q|<1. Symmetry in a and b sorts that out, but the paper should say so.

Bottom line: the Entry 12 proofs hold up, and they are worth having. Theorem 4.1 is likely correct, but as written it is a theorem statement with a proof sketch, not a completed proof. A referee can send this back for a precise domain statement and a real convergence argument. I would take it for peer review, but I would not cite Theorem 4.1 in its current form.

Referee Report

5 major / 4 minor

Summary. This paper treats Ramanujan's Entry 12, the identity (1.1) equating a ratio of q-products with an infinite continued fraction. The authors give two proofs. Section 2 uses Euler's method to derive an exact finite continued-fraction identity (2.3) together with the recursion (2.6), thereby establishing modified convergence. Section 3 uses the associated J-fraction H(x), generating functions, and Darboux's method to prove Theorem 3.1 and then Corollary 3.2, the Entry 12 identity. Section 4 generalizes H(x) to an x-dependent J-fraction, and Theorem 4.1 asserts an explicit evaluation X(x)=2ρF(ρ)/G(ρ) in terms of two 2φ1 series for all complex x outside (-1,1), except possibly finitely many points.

Significance. Entry 12 has a notoriously complicated proof history; providing two new, methodologically different proofs is a useful service. The Euler-method proof in Section 2 is explicit and yields exact finite identities, which is a strength. The Darboux proof in Section 3 is in the standard toolkit but is only sketched. The generalization in Section 4 is the only genuinely new result and is potentially interesting, but its proof is not yet complete: the parameter domain is not specified precisely, the cited Markov-theorem argument does not support the stated convergence domain, and the 2φ1 series can diverge for parameters allowed by the hypotheses. If the domain issues are resolved, the explicit evaluation would be a meaningful contribution.

major comments (5)

[Section 4, Eq. (4.1)] The scaled recurrence (4.1) does not follow from (1.3) with the stated choice η^2 = -4ab/(1-ab)^2. Substituting z = ηx into (1.3) and writing P_k(x) = D_k(ηx)/(η^k(1-ab)^k) gives P_{k+1}(x) = xP_k(x) + (q^{2k}/η)P_k(x) + [ab/(η^2(1-ab)^2)](1-bq^{2k-1}/a)(1-aq^{2k-1}/b)P_{k-1}(x). With the stated η^2, the coefficient of P_{k-1} is -1/4(1-bq^{2k-1}/a)(1-aq^{2k-1}/b), whereas (4.1) has +1/4. Thus the β_k in (4.2) has the opposite sign, so the positive-measure orthogonality conclusion from [9, Th. 2.5.2] is not justified. The stated value of c also appears inconsistent with η: from η^2 = -4ab/(1-ab)^2 one expects c = -1/η = -(1-ab)/(2√(-ab)) up to sign, not -(1-ab)^2/√(-ab). Please correct or verify these constants; they propagate into the definition of γ1,γ2 and hence into the final formula.
[Section 4, Theorem 4.1] The 2φ1 series F(ρ) and G(ρ) in Theorem 4.1 have arguments bq/a and b/(aq). The theorem states no bounds on these ratios. For a=0.1, b=-5, q=0.5, |ab|=0.5<1 and a,b have opposite signs, but |bq/a|=25>1 and |b/(aq)|=100>1, so both series diverge. The theorem is not well-defined as stated; either add convergence conditions (e.g., |b/a|<1) or explain continuation via the reciprocal symmetry analogous to Section 3's Remark (1).
[Section 4, Remark (3)] The claim that Markov's theorem implies convergence of X(x) for all x not in (-1,1) except possibly a finite set is unsupported. If the orthogonality measure has a discrete mass at M>1, the true interval of orthogonality contains [-1,M]; Markov's theorem [9, Th. 2.6.2] gives convergence only outside that interval, not on the gap (1,M). The paper's own Remark (2) admits discrete masses outside (-1,1). A separate argument covering the gaps, or a restriction excluding such masses, is required.
[Section 4, positivity of β_k] Even after correcting the sign in (4.1), the assertion that β_k>0 for all k follows from 'a and b have opposite signs' is not correct. β_k = 1/4(1-bq^{2k-1}/a)(1-aq^{2k-1}/b) is real positive only for suitable q (e.g., 0<q<1 real); for complex q or q<0 with |b/a| large the factors can be nonpositive. The theorem does not state that q is real. If the intended domain is real q in (0,1), this must be said, and the analytic continuation in q must be justified separately.
[Section 3, Theorem 3.1 and Corollary 3.2] Corollary 3.2 is stated under |q|<1, |ab|<1, but its proof uses Theorem 3.1, whose hypotheses include |a^2 q|<1, and the later use of (2.5)-(2.6) similarly requires |a^2 q|<1. No symmetry reduction is provided; since the identity (1.1) is symmetric in a and b, one can assume |a|<1 (or swap a,b) when |a|≥1, but the paper does not say this. As written, the proof does not cover, for example, a=2, b=0.1, q=0.5. Please add the reduction or state the restricted domain.

minor comments (4)

[Section 3, generating functions] The generating function formulas for ˆN(t) and ˆD(t) are stated without derivation; since Darboux's method depends on them, please include a derivation or a clear reference to the standard recurrence-to-generating-function calculation.
[Section 3, proof of Corollary 3.2] The phrase 'from (3)' should read 'from (3.2)'; no equation (3) exists in the paper.
[Section 1, display (1.2)] The J-fraction notation in (1.2) is ambiguous; the authors should indicate the continued-fraction separator convention, for example by writing it in the form A0/(A0x+B0 - C1/(A1x+B1 - ...)) as in [9, Ch. 2].
[Throughout] There are minor typographical errors, including 'birthda y' in the dedication, 'W e' at the start of the abstract, and 'F akult¨at f ¨ur' in the affiliation.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: Ramanujan's Entry 12 and the Section 4 generalization are derived from explicit q-series identities, generating functions, Darboux asymptotics, and external orthogonality theorems, without reusing the target identities as inputs.

full rationale

The paper's central claim (1.1) is proved in two independent ways. Section 2 starts from the q-binomial expansion (2.4) and derives the finite identity (2.3) by repeated applications of the elementary identity (2.1); the target continued fraction appears as its limiting or modified-convergence form, not as a premise. Section 3 derives H(1) by constructing generating functions for the numerator and denominator polynomials, applying Darboux's method and Heine's transformation, and Corollary 3.2 then uses the independently derived identities (2.5), (2.6), and (2.4) to convert H(1) into the infinite product. The Section 4 generalization X(x) is computed in Theorem 4.1 from the recurrence (4.1), the q-difference equation for Q(t), and Darboux's method, with convergence justified by external results from Blumenthal, Nevai, and Markov rather than by assuming the value to be proved. The self-citations [4] and [5] are used for historical context and as examples of similar computations; they are not premises of Theorem 3.1, Corollary 3.2, or Theorem 4.1, and neither target identity is used as an input anywhere. Thus the derivation is self-contained modulo standard background, and no circular step is present.

Assumptions & free parameters 0 free parameters · 9 assumptions · 0 invented entities

The proof rests on standard q-series and orthogonal polynomial theorems, plus the stated convergence and sign conditions. No free parameters are fitted; a, b, and q are free variables in the identity. The only invented mathematical objects are the continued fractions H(x) and X(x) themselves, which are explicitly defined and evaluated, not postulated without evidence.

assumptions (9)

standard math q-binomial theorem (Gasper and Rahman, Eq. I.3)
Used in Section 2, Eq. (2.4), to expand the product ratio as a quotient of two 2φ1 series. This is a standard identity in q-series.
standard math Heine's transformation for 2φ1 series (Gasper and Rahman, Eq. III.1)
Used in the proof of Theorem 3.1 to transform the asymptotic hypergeometric expressions into the stated form. Standard basic hypergeometric transformation.
standard math Darboux's method (Ismail, Th. 1.2.4)
Used in Sections 3 and 4 to obtain asymptotic formulas for numerator and denominator polynomials from their generating functions. The method is standard in asymptotic analysis of orthogonal polynomials.
standard math J-fraction to three-term recurrence correspondence (Ismail, Th. 2.6.1)
Used to derive the recurrence relation (1.3) from the J-fraction (1.2). Standard theory of continued fractions.
standard math Markov's theorem (Ismail, Th. 2.6.2)
Invoked in Section 4 to identify the limit of the polynomial ratio with the Stieltjes transform of the orthogonality measure. Standard in the theory of orthogonal polynomials.
standard math Positivity of recurrence coefficients implies orthogonality with respect to a positive measure (Ismail, Th. 2.5.2)
Used in Section 4 to conclude that the P_k(x) are orthogonal with respect to a positive measure for real a and b of opposite signs. Standard.
standard math Blumenthal's theorem and Nevai's theorem
Used in the remarks of Section 4 to describe the orthogonality measure; details are deferred. Standard results in orthogonal polynomial theory.
domain assumption Convergence conditions |q|<1, |ab|<1, and |a^2 q|<1 hold where stated
These conditions are stated in Theorem 3.1 and Corollary 3.2. They guarantee convergence of the q-products, the 2φ1 series, and the continued fraction.
domain assumption a and b are real and of opposite signs for the measure-theoretic part
Stated in Section 4 before (4.1). This makes c real and the recurrence coefficients β_k positive, so the denominator polynomials are orthogonal with respect to a positive measure; the proof of Theorem 4.1 relies on this.

how reviews work

0 comments

Cite this review

Pith. "Pith review of On Entry II.16.12: A continued fraction of Ramanujan." pith.science (2026). https://pith.science/paper/OO3J7HZG

@misc{pith2026190803333,
  author       = {Pith},
  title        = {Pith review of: On Entry II.16.12: A continued fraction of Ramanujan},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OO3J7HZG}},
  note         = {Machine review of arXiv:1908.03333}
}

read the original abstract

We study a continued fraction due to Ramanujan, that he recorded as Entry 12 in Chapter 16 of his second notebook. It is presented in Part III of Berndt's volumes on Ramanujan's notebooks. We give two alternate approaches to proving Ramanujan's Entry 12, one using a method of Euler, and another using the theory of orthogonal polynomials. We consider a natural generalization of Entry 12 suggested by the theory of orthogonal polynomials.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages

[4]

Bhatnagar

G. Bhatnagar. How to prove Ramanujan’s q-continued fractions. In Ramanujan 125, volume 627 of Contemp. Math. , pages 49–68. Amer. Math. Soc., Providence, RI, 2014

work page 2014
[5]

Bhatnagar and M

G. Bhatnagar and M. E. H. Ismail. Orthogonal polynomials associated with a continued fraction of Hirschhorn. arXiv e-prints , 22 pp., Jan 2019

work page 2019
[1]

Adiga, B

C. Adiga, B. C. Berndt, S. Bhargava, and G. N. W atson. Chap ter 16 of Ramanujan’s second notebook: theta-functions and q-series. Mem. Amer. Math. Soc. , 53(315):v+85, 1985

work page 1985
[2]

G. E. Andrews and B. C. Berndt. Ramanujan ’s lost notebook. Part II . Springer, New York, 2009

work page 2009
[3]

B. C. Berndt. Ramanujan ’s notebooks. Part III. Springer-Verlag, New York, 1991

work page 1991
[6]

T. S. Chihara. An introduction to orthogonal polynomials . Gordon and Breach Science Pub- lishers, New York-London-Paris, 1978. Mathematics and its Applications, Vol. 13

work page 1978
[7]

L. Euler. De transformatione seriei divergentis 1 − mx + m(m + n)x2 − m(m + n)(m + 2n)x3 + m(m + n)(m + 2 n)(m + 3 n)x4+ etc. in fractionem continuam. Nova Acta Academiae Scientarum Imperialis Petropolitinae , 2:36–45, 1788. Reprinted in Opera Om- nia: Series 1, Volume 16, pp. 34–46; English translation (by C. E. Sandifer) available at http://www.math.dar...

work page
[8]

Gasper and M

G. Gasper and M. Rahman. Basic Hypergeometric Series , volume 96 of Encyclopedia of Mathematics and its Applications . Cambridge University Press, Cambridge, second edition,

work page

Show all 16 references

[9]

M. E. H. Ismail. Classical and quantum orthogonal polynomials in one variab le, volume 98 of Encyclopedia of Mathematics and its Applications . Cambridge University Press, Cambridge, RAMANUJAN’S ENTRY II.16.12 15
[10]

Jacobsen

L. Jacobsen. Domains of validity for some of Ramanujan’ s continued fraction formulas. J. Math. Anal. Appl. , 143(2):412–437, 1989

1989
[11]

Lorentzen and H

L. Lorentzen and H. W aadeland. Continued fractions with applications , volume 3 of Studies in Computational Mathematics . North-Holland Publishing Co., Amsterdam, 1992

1992
[12]

P. G. Nevai. Orthogonal polynomials. Mem. Amer. Math. Soc. , 18(213):v+185, 1979

1979
[13]

K. G. Ramanathan. Hypergeometric series and continued fractions. Proc. Indian Acad. Sci. Math. Sci. , 97(1-3):277–296 (1988), 1987

1988
[14]

Ramanujan

S. Ramanujan. Collected papers of Srinivasa Ramanujan . AMS Chelsea Publishing, Provi- dence, RI, 2000. Edited by G. H. Hardy, P. V. Seshu Aiyar and B. M. Wilson, Third printing of the 1927 original, With a new preface and commentary by Bru ce C. Berndt. F akult¨at f ¨ur Mathe...

2000
[2004]

With a foreword by Richard Askey
[2009]

Askey, Reprint of the 2005 original

With two chapters by W alter Van Assche, With a foreword by Richard A. Askey, Reprint of the 2005 original

2005

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [4]

Bhatnagar

G. Bhatnagar. How to prove Ramanujan’s q-continued fractions. In Ramanujan 125, volume 627 of Contemp. Math. , pages 49–68. Amer. Math. Soc., Providence, RI, 2014

work page 2014

[2] [5]

Bhatnagar and M

G. Bhatnagar and M. E. H. Ismail. Orthogonal polynomials associated with a continued fraction of Hirschhorn. arXiv e-prints , 22 pp., Jan 2019

work page 2019

[3] [1]

Adiga, B

C. Adiga, B. C. Berndt, S. Bhargava, and G. N. W atson. Chap ter 16 of Ramanujan’s second notebook: theta-functions and q-series. Mem. Amer. Math. Soc. , 53(315):v+85, 1985

work page 1985

[4] [2]

G. E. Andrews and B. C. Berndt. Ramanujan ’s lost notebook. Part II . Springer, New York, 2009

work page 2009

[5] [3]

B. C. Berndt. Ramanujan ’s notebooks. Part III. Springer-Verlag, New York, 1991

work page 1991

[6] [6]

T. S. Chihara. An introduction to orthogonal polynomials . Gordon and Breach Science Pub- lishers, New York-London-Paris, 1978. Mathematics and its Applications, Vol. 13

work page 1978

[7] [7]

L. Euler. De transformatione seriei divergentis 1 − mx + m(m + n)x2 − m(m + n)(m + 2n)x3 + m(m + n)(m + 2 n)(m + 3 n)x4+ etc. in fractionem continuam. Nova Acta Academiae Scientarum Imperialis Petropolitinae , 2:36–45, 1788. Reprinted in Opera Om- nia: Series 1, Volume 16, pp. 34–46; English translation (by C. E. Sandifer) available at http://www.math.dar...

work page

[8] [8]

Gasper and M

G. Gasper and M. Rahman. Basic Hypergeometric Series , volume 96 of Encyclopedia of Mathematics and its Applications . Cambridge University Press, Cambridge, second edition,

work page