Asymptotics of Two Integrals from Optimization Theory and Geometric Probability

Author(s): D. J. Gates
Source: Advances in Applied Probability, Vol. 17, No. 4 (Dec., 1985), pp. 908-910
Published by: Applied Probability Trust
Stable URL: http://www.jstor.org/stable/1427094
Accessed: 05/11/2010 04:51

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=apt.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Applied Probability Trust is collaborating with JSTOR to digitize, preserve and extend access to Advances in
Applied Probability.

http://www.jstor.org
 Adv. Appl.Prob.17, 908-910 (1985)
Printedin N. Ireland
? AppliedProbabilityTrust1985

ASYMPTOTICS OF TWO INTEGRALS FROM OPTIMIZATION

THEORY AND GEOMETRIC PROBABILITY

D. J. GATES,* CSIRO Division of Mathematics and Statistics

Abstract
Asymptotic series are derived for two integrals using a Gaussian
identity and Laplace's method, demonstrating an improvement over
earlier methods.
LAPLACE'S METHOD; OPTIMIZATION

Anderssen et al. (1976) obtain various bounds and approximations for the expected
distance

(1) mk ? dXk
+ + X ) dxl
from the origin of a point uniformly distributed in the cube [0, 1]k. They evaluate m1,
m2 and m3 exactly. Otherwise their computationally most efficient formula, by far, is the
asymptotic series
(2) mk = (k/3)?(1- 1/10k- 13/280k2- 101/2800k3-37533/1232000k4)+ O(k )
as k -> oo.Terms up to k-3 give, for example, n4 accurate to five figures, m0lo accurate
to six figures and n20 accurate to seven figures. Their derivation of (2) is, however,
cumbersome. We give a simple derivation based on Laplace's method.
The authors also study the expected interpoint distances

(3) Mk = -.. {x,- y+)2 +(xyk)2}1dx idy. dxk dYk,

but do not give an asymptotic series like (2), presumably because of the work required
using their method. We give a simple derivation of such a series, again using Laplace's
method.
Since

(4) AX=(2/V)i)Ak ds exp (- 1s2)

we can write

(5) = (2/rr)kJ f'(- s2)(- S2k-1 ds,

mk

Received 28 August 1985.

* Postal address: CSIRO Division of Mathematics and
Statistics, GPO Box 1965, Canberra
ACT 2601, Australia.

908
Lettersto the editor 909

where

(6) f(t) = fexp (tx2) dx.

Since f(t) has a maximum at t = 0, and f'(0)= 3, we write

f(t)= exp (t/3) {1+ dx

(7) t(x2-)+t2(2-)2 .}
= exp (t/3)(1+ 2t2/45+
??.).
Similarly
(8) f'(t) = exp (t/3)( + 4t/45 + * ).
Finally
!-
mk = (2/,rr)kJ exp (-ks2/6)( s2+-ks+ ...) ds
(9)
= "
)(1- 1/10k + ).

Turning now to (3) we have, similarly,

(10) Mk = S,
(2/w)kg'(-s2)g(-2k1
where

(11) g(t) = J1exp (t(x - y)2) dx dy,

which has a maximum at t = 0, where g'= g. Thus we write

(12) g(t)= exp(t/6)

n=O t"In.
where

In = -
I{(x-yy--}dxdy,
and

(13) g'(t) = exp (t/6) t"J,,

n=O

where

J, = In/6 + (n + 1)In,1.
Then Io=1, I1=0, 12=7/360, 13= 11/5670, Jo= 1/6, J1=7/180 and J2= 137/15120.
Now putting (12) and (13) in (10) and using standard formulae for moments of a
normal density gives

(14) Mk = (k/6)A(1-7/40k -65/896k2+

.).
Anderssen et al. (1976) compute M1, M2 exactly and , Mlo by a slowly
M3,'"
910 Lettersto theeditor

convergent series method. They also obtain an upper bound

(15) Mk < (k/6)1[{1+ 2(1- 3/5k)'}/3]1.
The table lists the M1, , from Anderssen et al. and their deviations from (14) (as
shown) and (15) denoted ? Ml,,
(14)-Mk and (15)-Mk respectively. This illustrates the
accuracy of (14), for k not too small, while its efficiency is obvious.

k Mk (14)- Mk (15) - Mk

1 0-33333 -0-026 0-021

2 0-52141 -0-005 0-024
3 0-66167 -0-001 0-020
4 0-77766 -0-0006 0-017
5 0-87853 -0-0003 0-015
6 0-96895 -0-0001 0-014
7 1-05159 -0-00007 0-013
8 1-12817 -0-00004 0-012
9 1-19985 -0-00002 0.011
10 1-26748 -0-00001 0.010

References
ANDERSSEN, R. S., BRENT, R. P., DALEY, D. J. AND MORAN,P. A. P. (1976) Concerning
f1... (x0 + + x)dxdx 1 dXk and a Taylor series method. SIAM J. Appl. Math. 30, 22-30.
..