Iterated Radical Expansions and Convergence
Abstract
We treat three recurrences involving square roots, the first of which arises from an infinite simple radical expansion for the Golden mean, whose precise convergence rate was made famous by Richard Bruce Paris in 1987. A never-before-seen proof of an important formula is given. The other recurrences are non-exponential yet equally interesting. Asymptotic series developed for each of these two examples feature a constant, dependent on the initial condition but otherwise intrinsic to the function at hand.
From the intricacies of nonlinear recurrences emerge a plethora of constants. Our work builds on what we began in [1, 2].
1 The Map
As , the recurrence
|
approaches the Golden mean [3]:
from below and enjoys exponential convergence [4]:
We shall prove this formula using entirely elementary techniques.
First, notice that for all by induction ( is obvious; supposing , we obtain
). Now, writing , we have , and
because
hence
hence
hence
thus
for all .
From , we observe
hence
hence
hence
hence
because ; therefore
exists and is nonzero since
converges. This completes the proof. The numerically-efficient expression for
as an infinite product did not appear in [4], but was subsequently found by Philippe Flajolet & Paul Zimmermann several years later [5]. No trace of their derivation has survived; the above proof is new. Other calculations of are exhibited in [6, 7].
2 The Map
Given the recurrence
we start with asymptotics [9]
valid as , for some constant . On the basis of numerical experimentation, we hypothesize that the next terms of the asymptotic series must be of the form
The challenge is to express each coefficient , , , as a polynomial in . To find these, we replace by everywhere:
and expand in powers of and :
To avoid dealing with the radical in
square both sides, obtaining
i.e.,
Upon rearrangement, relevant terms (in decreasing order of significance) of become
Performing an analogous substitution in , the corresponding terms of become
Matching coefficients, we obtain
which are consistent with [10] and
which are new (as far as is known).
These fourteen parameter values allow us to estimate the constant . Our simple procedure involves computing exactly via recursion, setting this equal to our series (up to ) and then solving:
Note that the estimate appears in [11]. We find the implicit representation
to be intriguing: the left-hand side echoes the logistic map but only somewhat: it is off by a sign. In the following section, a comparable implicit representation leads to a surprising outcome.
3 The Map
Given the recurrence
we start with conjectured asymptotics (with no known theoretical basis)
as , for some constant . It is reasonable to hypothesize that the next terms of this series are
To find , we replace by everywhere:
and expand in powers of and :
To avoid dealing with the radical in
square both sides, obtaining
i.e.,
i.e.,
Upon rearrangement, the terms of involving either or become most relevant. Performing an analogous substitution in , we match coefficients as before. The terms containing give
and the terms containing give
It is astonishing that we have seen these coefficients before – review our analysis [2] of the recurrence
which yields the identical seven expressions – although the signs preceding each coefficient may differ. This is very surprising! While the sequences behave distinctly, there is a hidden commonality in structure, captured by the polynomials in .
These seven parameter values allow us to estimate the constant . Our simple procedure involves computing exactly via recursion, setting this equal to our series (up to ) and then solving:
Note that the estimate appears in [12].
The seemingly arbitrary maps in Sections 2 & 3 turn out to be curiously linked: setting , we have
Their appearance on a digital bulletin board [11, 12] by different participants three years apart would suggest that there is no connection. The mystics among us, however, might insist that there are no coincidences.
4 Acknowledgements
My proof in Section 1 is original, but it follows the outline of a related proof that Robert Israel and Anthony Quas gave long ago (reproduced in [1]). I am thankful for helpful correspondence with Paul Zimmermann and Michael Somos. My attempts to reach Dumitru Popa [9, 10] have regrettably failed. In the statement of Theorem 5 on page 21 of [10], the lead coefficient of the term should be . This particular theorem provides us rigorously with the “seed” (initial terms) from which the asymptotic series in Section 2 grows. It does not cover the example in Section 3 – the reason is that the derivative – a generalization over-and-beyond the archetypal scenario will be needed to fill the gap.
5 Addendum:
We’ve discovered that the expressions
enjoy a certain kindredship, in the sense that iterations based on these functions possess strikingly similar asymptotic series. The same turns out to be true for
Consider the recurrence , for . We find
and, unlike the alternating blocks of positive & negative signs in Section 2, here all signs are positive (except for the earliest & terms). More on this phenomenon is given in [13]. Using the same procedure as before, we obtain
The commonality in structure between two functions does not assist in the numerical calculation of constants. No relationship between and is observed (nor was any expected).
References
- [1] S. R. Finch, A deceptively simple quadratic recurrence, arXiv:2409.03510.
- [2] S. R. Finch, Generalized logistic maps and convergence, arXiv:2409.15175.
- [3] S. R. Finch, The Golden mean, Mathematical Constants, Cambridge Univ. Press, 2003, pp. 5–11; MR2003519.
- [4] R. B. Paris, An asymptotic approximation connected with the golden number, Amer. Math. Monthly 94 (1987) 272–278; MR0883295.
- [5] S. Plouffe, The Paris constant, http://plouffe.fr/simon/constants/paris.txt; the numerical estimate of here is accurate only to 48 digits.
- [6] N. J. A. Sloane, Decimal expansion of the Paris constant (accurate to 100 digits), http://oeis.org/A105415.
- [7] M. Somos, contribution to thread about recursion , , http://math.stackexchange.com/questions/4508647/closed-form-of-a-n1-sqrta-n1-a-1-1/4508737#4508737.
- [8] J D.’Aurizio, contribution to thread about recursion , , http://math.stackexchange.com/questions/1304500/recurrence-relations-solving-for-b-n.
- [9] D. Popa, Recurrent sequences and the asymptotic expansion of a function, Gazeta Mat. Ser. A, v. 37 (2019) n. 3-4, 1–16; http://ssmr.ro/gazeta/gma/2019/gma3-4-2019-continut.pdf.
- [10] D. Popa, Refined asymptotic expansions for some recurrent sequences, Gazeta Mat. Ser. A, v. 41 (2023) n. 1-2, 18–26; http://ssmr.ro/gazeta/gma/2023/gma1-2-2023-continut.pdf.
- [11] M. Somos, contribution to thread about recursion , , http://math.stackexchange.com/questions/4245022/deriving-an-approximation-for-a-recursive-sequence/4359477#4359477.
- [12] M. Somos, contribution to thread about recursion , , http://math.stackexchange.com/questions/2914208/what-is-this-function-related-with-continued-fractions/2914395#2914395.
-
[13]
S. R. Finch, What do and
have in common? arXiv:2411.01591.
Steven Finch MIT Sloan School of Management Cambridge, MA, USA steven_finch_math@outlook.com