|
The "One-Ninth" Constant
We are concerned here with rational approximation of exp(-x) on the
half-line [0,). Let
denote the error of best uniform
approximation:
where
There are two cases of special interest, when m=0 and m=n,
since clearly
Many researchers [1,2,10] have studied these constants
, referred to as Chebyshev constants
in [2]. We mention the work of only a few. Schönhage [11] proved
that
which led several people to conjecture that
Numerical evidence uncovered by Schönhage [12] and Trefethen &
Gutknecht [13] suggested that the conjecture is false. Carpenter,
Ruttan and Varga [14] calculated the Chebyshev constants to an
accuracy of 200 digits up to n = 30 and carefully obtained
although a proof that the limit even existed was still to be
found.
Building on Opitz & Scherer [15] and Magnus [16], Gonchar & Rakhmanov [2,3] proved that the
limit exists and that it equals
where K is the complete elliptic integral of the first kind
and the constant c is defined as follows. Let E be the complete elliptic
integral of the second kind
(One's first encounter with K and E is often with regard to computing
the period of a physical pendulum and finding the arclength of an ellipse.)
The constant c is defined simply as the unique solution, 0 < c < 1, of
the equation
The appearance of elliptic integrals here reminds one of work done over one hundred years
before. In 1877, Zolotarev determined a number of exact solutions to
approximation problems using elliptic functions [1] in research
which was far ahead of its time. See Bernstein's constant
for more details.
Gonchar and Rakhmanov's exact disproof of the "one-ninth" conjecture
utilized ideas from complex potential theory, which seems far removed
from the rational approximation of exp(-x)! They also obtained a
number theoretic characterization of the "one-ninth" constant
. If
where
then f is complex-analytic in the open unit disc. The unique positive
root of the equation
is the constant . Another way
of writing
aj
is as follows [5]. If
is the prime factorization of the integer j, where each
pi
is an odd prime,
m ≥
0
and
mi ≥ 1
, then
Carpenter [2] computed to
101 digits using this equation.
Here is another expression due to Magnus [16]. The one-ninth constant
is the unique solution, with
0 < x < 1, of the equation
which turns out to have been studied one hundred years earlier by
Halphen [4]. Halphen was interested in theta functions and computed
to six digits, clearly unaware
that this constant would become prominent a century later! Varga [2] has
suggested that be renamed the
Halphen constant. So many researchers have contributed to the solution
of this approximation problem, however, that retaining the amusingly
inaccurate "one-ninth" designation might be simplest.
Relevant Mathcad files will be included as time permits.
Plouffe gave accurate approximations of
and 1/ in the
Inverse
Symbolic Calculator web pages.
Postscript
The constant c=0.9089085575... defining
arises in a completely unrelated field: the study of Euler elasticae [6-8].
A quotient of elliptic functions, similar to that above, occurs in our essay on
Grötzsch ring constants.
References
- P. P. Petrushev and V. A. Popov, Rational Approximation of Real
Functions, Cambridge Univ. Press, 1987;
MR 89i:41022.
- R. S. Varga, Scientific Computation on Mathematical Problems and
Conjectures, SIAM, 1990;
MR 92b:65012.
- A. A. Gonchar and E. A. Rakhmanov, Equilibrium distributions and degree
of rational approximation of analytic functions, Math. USSR Sbornik
62 (1989) 305-348;
MR 89h:30054.
- G.-H. Halphen, Traité
des fonctions elliptiques et de leurs applications,
v. 1, Gauthier-Villars, 1886, p. 287 (Cornell Digital Library).
- A. A. Gonchar, Rational approximations of analytic functions, Amer.
Math. Soc. Transl. Ser. 2, 147 (1990) 25-34;
MR 89e:30066.
- D. A. Singer, Curves whose curvature depends on distance from the origin,
Amer. Math. Monthly 106 (1999) 835-841;
MR 2000j:53005.
- T. A. Ivey and D. A. Singer, Knot types, homotopies and stability of closed
elastic rods, Proc. London Math. Soc. 79 (1999) 429-450;
MR 2000g:58015.
- C. Truesdell, The influence of elasticity on analysis: The classic heritage,
Bull. Amer. Math. Soc. 9 (1983) 293-310;
MR 85f:01004.
- R. S. Varga, Topics in Polynomial and Rational Interpolation and Approximation,
Les Presses de l'Université de Montréal, 1982;
MR 83h:30041.
- W. J. Cody, G. Meinardus and R. S. Varga, Chebyshev rational approximations to e-x in
[0,+) and applications to heat-conduction problems,
J. Approx. Theory 2 (1969) 50-65;
MR 39 #6536.
- A. Schönhage, Zur rationalen Approximierbarkeit von e-x über
[0,), J. Approx. Theory 7 (1973) 395-398;
MR 49 #3391.
- A. Schönhage, Rational approximation to e-x and related L2-problems,
SIAM J. Numer. Anal. 19 (1982) 1067-1080;
MR 83k:41016.
- L. N. Trefethen and M. H. Gutknecht, The Carathéodory-Fejér method for real rational approximation,
SIAM J. Numer. Anal. 20 (1983) 420-436;
MR 85g:41024.
- A. J. Carpenter, A. Ruttan, and R. S. Varga, Extended numerical computations on the "1/9" conjecture in rational
approximation theory, Rational Approximation and Interpolation, Lecture Notes in Math. 1105,
Springer-Verlag, 1984, pp. 383-411.
- H.-U. Opitz and K. Scherer, On the rational approximation of e-x on [0,),
Constr. Approx. 1 (1985) 195-216;
MR 88f:41027.
- A. P. Magnus, On the use of the Carathéodory-Fejér method for investigating "1/9" and similar
constants, Nonlinear Numerical Methods and Rational Approximation, Reidel, 1988, pp. 105-132,
MR 90j:65035.
- A. P. Magnus, Asymptotics and super asymptotics for best rational approximation error norms to the
exponential function (the "1/9" problem) by the Carathéodory-Fejér method, Nonlinear Numerical
Methods and Rational Approximation II, Kluwer, 1994, pp. 173-185;
preprint;
MR 96b:41023.
- A. P. Magnus and J. Meinguet, The elliptic functions and integrals of the "1/9" problem,
Numer. Algor. 24 (2000) 117-139 ; preprint.
Return to the Favorite Mathematical Constants main essay.
Copyright © 1995-2001 by Steven Finch.
All rights reserved.
|