Arithmetic properties of the Taylor coefficients of differentially algebraic power series
C. Krattenthaler † and T. Rivoal · 2025
Read Arithmetic properties of the Taylor coefficients of differentially algebraic power series by C. Krattenthaler † and T. Rivoal (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-09259.txt.
Introduction
Christian Krattenthaler and Tanguy Rivoal study the number patterns hidden in series solutions of differential equations, in a 2025 paper. A function defined by such an equation can be expanded as a power series, and its coefficients are built one by one through a nonlinear recurrence. A classic question asks how large the denominators of those fractions can grow.
Old bounds said the denominators could explode fast, but in the linear case they stay tame. The authors prove a strong bound for the nonlinear case too, whenever a key polynomial from the recurrence splits over the rational numbers. The denominators then divide a clean factorial-style expression. The proof uses detailed p-adic analysis. Examples include the Weierstrass elliptic function, Painlevé equation solutions, and Lagrange's solution of Kepler's equation.
Catalog record
| CREATOR | C. Krattenthaler † and T. Rivoal |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 70 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-09259
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Arithmetic properties of the Taylor coefficients of differentially algebraic power series Thanks: † Research partially supported by the Austrian Science Foundation FWF, grant 10.55776/F1002, in the framework of the Special Research Program “Discrete Random Structures: Enumeration and Scaling Limits”. C. Krattenthaler † and T. Rivoal Abstract . Let f = ∑ n = 0 ∞ f n x n ∈ ℚ ¯ [ [ x ] f=\sum_{n=0}^{\infty}f_{n}x^{n}\in\overline{\mathbb{Q}}[[x] be a solution of an algebraic differential equation Q ( x , y ( x ) , … , y ( k ) ( x ) ) = 0 Q(x,y(x),\ldots,y^{(k)}(x))=0 , where Q Q is a multivariate polynomial with coefficients in ℚ ¯ \overline{\mathbb{Q}} . The sequence ( f n ) n ≥ 0 (f_{n})_{n\geq 0} satisfies a non-linear recurrence, whose expression involves a polynomial M M of degree s s . When the equation is linear, M M is its indicial polynomial at the origin. We show that when M M is split over ℚ \mathbb{Q} , there exist two positive integers δ \delta and ν \nu such that the denominator of f n f_{n} divides δ n + 1 ( ν n + ν ) ! 2 s \delta^{n+1}(\nu n+\nu)!^{2s} for all n ≥ 0 n\geq 0 , generalizing a well-known property when the equation is linear. This proves in this case a strong form of a conjecture of Mahler that Pólya–Popken’s upper bound n 𝒪 ( n log ( n ) ) n^{\mathcal{O}(n\log(n))} for the denominator of f n f_{n} is not optimal. This also enables us to make Sibuya and Sperber’s bound | f n | v ≤ e 𝒪 ( n ) |f_{n}|_{v}\leq e^{\mathcal{O}(n)} , for all finite places v v of ℚ ¯ \overline{\mathbb{Q}} , explicit in this case. Our method is completely effective and rests upon a detailed p p -adic analysis of the above mentioned non-linear recurrences. Finally, we present various examples of differentially algebraic functions for which the associated polynomial M M is split over ℚ \mathbb{Q} , among which are Weierstraß’ elliptic ℘ \wp function, solutions of Painlevé equations, and Lagrange’s solution to Kepler’s equation. 2020 Mathematics Subject Classification Primary 11B37; Secondary 05A10 11B65 30B10 In the memory of Bernard Malgrange 1. Introduction We denote by \overline{\mathbb{Q}} the field of algebraic numbers and by \mathcal{O}_{\overline{\mathbb{Q}}} the ring of algebraic integers. Let us consider a non-trivial algebraic differential equation Q(x,y(x),\ldots,y^{(k)}(x))=0, (1.1) where Q\in\mathbb{C}[X,Y_{0},\ldots,Y_{k}] ; here and in the sequel, by “non-trivial” we mean that Q is of degree \geq 1 in at least one of the indeterminates Y_{j} . Let us assume that ( 1.1 ) has a power series solution f(x):=\sum_{n=0}^{\infty}f_{n}x^{n}\in\mathbb{C}[[x]] . We say that f is a differentially algebraic power series, and write DA in short. For instance, the Taylor expansions at x=0 of x/\log(1+x) and of \tan(x) are DA power series, as these functions are solutions of the equations x(1+x)y^{\prime}+y^{2}-(1+x)y=0 and y^{\prime}-y^{2}-1=0 , respectively, which are both Riccati equations. A natural question arises: is it possible to bound |f_{n}| non-trivially from above and from below? In the Archimedean case, an upper bound was first obtained by Maillet [ 16 ] , and made more precise by various authors, in particular by Malgrange [ 17 ] , and in the linear case by Perron [ 18 ] . It is recalled in [ 15 , p. 200] in the following form: there exist \alpha,\beta>0 that depend on Q and f such that |f_{n}|\leq\alpha n!^{\beta} for all n\geq 0 . A lower bound was obtained by Popken [ 20 ] under the additional assumption that f_{n}\in\overline{\mathbb{Q}} for all n\geq 0 , when \overline{\mathbb{Q}} is embedded into \mathbb{C} : there exist \alpha,\beta>0 that depend on Q and f such that either f_{n}=0 or |f_{n}|\geq\alpha n^{-\beta n\log(n)} for all n\geq 1 , and if Q is a linear polynomial in Y_{0},\ldots,Y_{k} ( i.e. , the differential equation is linear inhomogeneous with coefficients in \mathbb{C}[X] ), then either f_{n}=0 or |f_{n}|\geq\alpha n^{-\beta n} for all n\geq 1 . Similar lower bounds had been obtained earlier by Pólya [ 19 ] when f_{n}\in\mathbb{Q} . Popken’s lower bound has applications in the Theory of Transcendental Numbers, in particular to the values of Weierstraß’ \wp function; see [ 15 , pp. 207–212] . It was reproved in detail by Mahler in [ 15 , Chapter 8] , following the method used by Pólya [ 19 ] in the rational case. There it is shown to be a simple consequence of the following result. Let d_{n} denote the least positive integer such that d_{n}f_{n} is in \mathcal{O}_{\overline{\mathbb{Q}}} . We shall from now on say that d_{n} is “the denominator of” f_{n} . Then there exist \alpha,\beta>0 that depend on Q and f such that d_{n}\leq\alpha n^{\beta n\log(n)},\quad n\geq 1, (1.2) and if Q is a linear polynomial in Y_{0},\ldots,Y_{k} then d_{n}\leq\alpha n^{\beta n} for all n\geq 1 . Popken’s theorem follows by combining Maillet’s upper bound for the Galois conjugates \sigma(f_{n}) of f_{n} (because \sum_{n}\sigma(f_{n})x^{n} is also DA) and the upper bound ( 1.2 ) for d_{n} , and then by considering the norm of d_{n}f_{n} over \mathbb{Q} , which is a positive integer when f_{n}\neq 0 . Moreover, Mahler [ 15 , p. 212] conjectured in rather loose terms that the exponent \beta n\log(n) in Popken’s lower bound “can probably be improved to something like” {\beta}n\log\log(n) . This would make the general bound closer to the bound in the linear case. Such an improvement, which could be difficult to detect numerically, would follow immediately from the same improvement on the exponent of n! in ( 1.2 ) because in Maillet’s upper bound the exponent is even \mathcal{O}(1) . A different type of improvement of ( 1.2 ) was made by Sibuya and Sperber in [ 24 , p. 112, Eq. (III’)] , who proved that, for all finite places v of \overline{\mathbb{Q}} , |f_{n}|_{v}\leq e^{cn} for some non-explicit constant c>0 that depends on f , Q , and v ; this had been conjectured by Dwork. This proves in particular that f has positive v -adic radius of convergence in \mathbb{C}_{p} . Earlier, Mahler [ 14 ] had proved a lower bound for |f|_{v} similar to Popken’s in the Archimedean case. We refer to [ 23 ] for a survey of Archimedean and non-Archimedean estimates for f_{n} when f is a DA power series, in particular for (possibly not optimal) explicit values of the constants \alpha and \beta . Mahler’s denominator conjecture was the starting point of our investigations, and this also led us naturally to study Sibuya and Sperber’s v -adic bounds. Let us first consider two simple examples in the linear case, both of hypergeometric type. The series h_{1}(x):=\sum_{n=0}^{\infty}\frac{x^{k}}{\prod_{k=0}^{n-1}(k^{2}+1)} is a DA power series solution to x^{2}y^{\prime\prime\prime}+xy^{\prime\prime}+(1-x)y^{\prime}-y=0 , and the associated denominator d_{n} satisfies d_{n}=\prod_{k=0}^{n-1}(k^{2}+1)\leq\prod_{k=0}^{n-1}(k+1)^{2}=n!^{2} , as expected. Notice however that, conjecturally, there exist no integers \delta,\nu,\mu,s\geq 0 such that d_{n} divides \delta^{n+1}(\nu n+\mu)!^{s} for all n\geq 0 ( 1 1 1 Assume on the contrary the existence of such integers. Then, any prime divisor p of d_{n} divides either \delta or (\nu n+\mu)! , hence p\leq\max(\delta,\nu n+\mu) . But it is widely believed, because it is an instance