A real random variable
X is sub-Gaussian iff there exist
K>0 such that
E[exp(λ
X)]≤exp(??) for any λ ∈
RJ-P. Kahane [10] proved that a real random variable
X is sub-Gaussian if and only if
E[
X]=0 and
E[exp(
εX2)] <∞ for some
ε>0.
A probability measure
μ on a Banach space
B is said to be
sub-Gaussian iff there exists
C>0 such that
∫
Bexp(<
y, x>)
μ(
dx)≤exp(??∫<
y, x>
2μ(
dx))<∞ for any
y ∈
B *. (1)
A Gaussian measure and the probability measure induced by a Rademacher series are typical examples of sub-Gaussian measures, and for these two probability measures, exp(
ε||
x||
2) is integrable for some
ε>0 ([3], [12]). We call this integrability the
exponential square integrability. When
B =
L P (
p≥1), (1) is a sufficient condition for the exponential square integrability, but not necessary even if
B is a Hilbert space ([4]).
For a sub-Gaussian measure
μ, the
Lp (
μ) topologies (0 <
p < ∞) and the
L0(
μ) topology coincide on the family {<
y, x> ;
y ∈
B*}. To show that considerably many probability measures satisfy such a remarkable property, we shall propve the sub-Gaussian property of probability measures for two types. One is a probability measure identified with a positive generalized Wiener function (see H. Sugita [17]), and the other is a probability measure which is absolutely continuous with respect to the probability measure induced by a random Fourier series. The former is exponentially square integrable [17], and so is the latter under certain additional conditions.
抄録全体を表示