PROCEEDINGS OF THE
AMERICAN MATHEMATICALSOCIETY
Volume 109, Number 2, June 1990
THE DISTRIBUTION OF RADEMACHER SUMS
S. J. MONTGOMERY-SMITH
(Communicated by William D. Sudderth)
Abstract. We find upper and lower bounds for Pr(23 +-Xn > t), where x, ,
x2, ... are real numbers. We express the answer in terms of the ^-interpolation
norm from the theory of interpolation of Banach spaces.
Introduction
Throughout this paper, we let e{, e2, ... be independent Bernoulli random
variables (that is, Pr(e/J = 1) = Pr(en = -1) = \). We are going to look for
upper and lower bounds for Pr(£] enxn > t), where xx, x2, ... is a sequence
of real numbers such that x = (xn)^=x e l2.
Our first upper bound is well known (see, for example, Chapter II, §59 of
[5]):
d) pr(Ee^>'W2)<^'2/2-
However, if ||x||, < oo, this cannot also provide a good lower bound, because
then we have another upper bound:
(2) Pr(EV->IMIi)-0-
To look for lower bounds, we might first consider using some version of the
central limit theorem. For example, using Theorem 7.1.4 of [2], it can be shown
that for some constant c we have
Thus, for some constant c we have that if r < c~ (log||x||3/||x||2) , then
/•oo 2 ~2 _'2/2
Pr(£e^>'Nl2)>^7, e~s/2ds>C-^-t-.
Received by the editors December 22, 1988 and, in revised form, August 30, 1989.
1980 Mathematics Subject Classification (1985 Revision). Primary 60C05; Secondary 60G50.
Key words and phrases. Rademacher sum, Holmstedt's formula.
©1990 American Mathematical Society
0002-9939/90 $1.00+ $.25 per page
517
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use