06.06.2013 Views

Theory of Statistics - George Mason University

Theory of Statistics - George Mason University

Theory of Statistics - George Mason University

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

734 0 Statistical Mathematics<br />

**************domain *** real line versus complex plane ******************<br />

refer to pro<strong>of</strong> <strong>of</strong> Theorem 1.17) on page 49 as an example <strong>of</strong> analytic continuation.<br />

The property <strong>of</strong> being analytic is quite different for real and complex<br />

functions. In the case <strong>of</strong> a complex function <strong>of</strong> a complex variable f(z), if the<br />

first derivative <strong>of</strong> f exists at all points within a region D, then the derivatives<br />

<strong>of</strong> all orders exist. Furthermore, the Taylor series converges to the function<br />

value within the region over which the function is analytic. (These facts can<br />

be shown using the Cauchy integral formula; see Churchill, 1960, Chapters 5<br />

and 6, for example.) The definition <strong>of</strong> an analytic complex function is usually<br />

different from that <strong>of</strong> a real function. An analytic complex function is defined<br />

as one whose (first) derivative exists over a region.<br />

0.1.9 L p Real Function Spaces<br />

Definition 0.1.42 (L p space)<br />

Given the measure space (Ω, F, ν) and the real number p ≥ 1. The space <strong>of</strong><br />

all measurable functions f on Ω for which |f| p dν < ∞ is called the L p (ν)<br />

space, or just the L p space.<br />

Although the measure ν is needed to define the integral, we <strong>of</strong>ten drop the<br />

ν in L p (ν). If the integral is taken only over some D ∈ F, we may denote the<br />

space as L p (D), and a more complete notation may be L p (ν, D).<br />

An L p space is a linear space (exercise).<br />

An important fact about the L p spaces is that they are Banach spaces<br />

(that is, among other things, they are complete). This fact is called the Riesz-<br />

Fischer theorem and is proved in most texts on real analysis.<br />

There are several types <strong>of</strong> useful operations on functions in a given function<br />

space F. Most binary operations require that the domains <strong>of</strong> the two functions<br />

be the same. Function composition, equation (0.1.68), is a very common<br />

operation that only requires that the range <strong>of</strong> one function be in the domain <strong>of</strong><br />

the other. Many interesting binary operations on functions involve integration<br />

<strong>of</strong> the functions, and so require that the functions be in some L p space.<br />

We now describe some <strong>of</strong> these operations. Each operation is a mapping.<br />

Some binary operations map L p × L p to L q or map L p × L p to IR, <strong>of</strong>ten for<br />

p = 2. Some useful unary operations map L p to IR, to [−1, 1], or to ĪR+. The<br />

transforms described in Section 0.1.12 map L 1 to L 1 .<br />

0.1.9.1 Convolutions and Covariances and Correlations<br />

The convolution <strong>of</strong> the functions f and g is<br />

<br />

(f ⋆ g)(t) = f(x)g(t − x)dx. (0.1.69)<br />

<strong>Theory</strong> <strong>of</strong> <strong>Statistics</strong> c○2000–2013 James E. Gentle<br />

D

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!