In probability theory, the Wick product is a particular way of defining an adjusted product of a set of random variables. In the lowest order product the adjustment corresponds to subtracting off the mean value, to leave a result whose mean is zero. For the higher order products the adjustment involves subtracting off lower order (ordinary) products of the random variables, in a symmetric way, again leaving a result whose mean is zero. The Wick product is a polynomial function of the random variables, their expected values, and expected values of their products.
The definition of the Wick product immediately leads to the Wick power of a single random variable and this allows analogues of other functions of random variables to be defined on the basis of replacing the ordinary powers in a power-series expansions by the Wick powers. The Wick powers of commonly-seen random variables can be expressed in terms of special functions such as Bernoulli polynomials or Hermite polynomials.
The Wick product is named after physicist Gian-Carlo Wick, cf. Wick's theorem.
Assume that X1, ..., Xk are random variables with finite moments. The Wick product
\langleX1,...,Xk\rangle
is a sort of product defined recursively as follows:
\langle\rangle=1
(i.e. the empty product - the product of no random variables at all - is 1). For k ≥ 1, we impose the requirement
{\partial\langleX1,...,Xk\rangle\over\partialXi} =\langleX1,...,Xi-1,\widehat{X}i,Xi+1,...,Xk\rangle,
where
\widehat{X}i
\operatorname{E}\langleX1,...,Xk\rangle=0.
Equivalently, the Wick product can be defined by writing the monomial
X1...Xk
X1...Xk=\sumS\subseteq\left\{1,...,k\right\
where
\langleXi:i\inS\rangle
\langle
X | |
i1 |
,...,X | |
im |
\rangle
S=\left\{i1,...,im\right\}
It follows that
\langleX\rangle=X-\operatorname{E}X,
\langleX,Y\rangle=XY-\operatorname{E}Y ⋅ X-\operatorname{E}X ⋅ Y+2(\operatorname{E}X)(\operatorname{E}Y)-\operatorname{E}(XY),
\begin{align} \langleX,Y,Z\rangle =&XYZ\\ &-\operatorname{E}Y ⋅ XZ\\ &-\operatorname{E}Z ⋅ XY\\ &-\operatorname{E}X ⋅ YZ\\ &+2(\operatorname{E}Y)(\operatorname{E}Z) ⋅ X\\ &+2(\operatorname{E}X)(\operatorname{E}Z) ⋅ Y\\ &+2(\operatorname{E}X)(\operatorname{E}Y) ⋅ Z\\ &-\operatorname{E}(XZ) ⋅ Y\\ &-\operatorname{E}(XY) ⋅ Z\\ &-\operatorname{E}(YZ) ⋅ X\\ &-\operatorname{E}(XYZ)\\ &+2\operatorname{E}(XY)\operatorname{E}Z+2\operatorname{E}(XZ)\operatorname{E}Y+2\operatorname{E}(YZ)\operatorname{E}X\\ &-6(\operatorname{E}X)(\operatorname{E}Y)(\operatorname{E}Z). \end{align}
In the notation conventional among physicists, the Wick product is often denoted thus:
:X1,...,Xk:
and the angle-bracket notation
\langleX\rangle
is used to denote the expected value of the random variable X.
The nth Wick power of a random variable X is the Wick product
X'n=\langleX,...,X\rangle
with n factors.
The sequence of polynomials Pn such that
Pn(X)=\langleX,...,X\rangle=X'n
form an Appell sequence, i.e. they satisfy the identity
Pn'(x)=nPn-1(x),
for n = 0, 1, 2, ... and P0(x) is a nonzero constant.
For example, it can be shown that if X is uniformly distributed on the interval [0, 1], then
X'n=Bn(X)
where Bn is the nth-degree Bernoulli polynomial. Similarly, if X is normally distributed with variance 1, then
X'n=Hn(X)
where Hn is the nth Hermite polynomial.
(aX+bY)'n=
n | |
\sum | |
i=0 |
{n\choosei}aibn-iX'iY'{n-i
\langle\operatorname{exp}(aX)\rangle \stackrel{def