A Hilbert space is a generalized complex vector space , which may have an (uncountably) infinite number of dimensions, and on which the following inner product is defined: For any and , we define their inner product such that
i) ,
ii) , and
iii) , and if and only if .
Furthermore, we require that is complete with respect to the norm . Let be a sequence such that for every real number , there exists an integer such that for all integers ,
.
Then the sequence is called Cauchy, and the completeness axiom states that every Cauchy sequence of vectors in converges to a vector in .
Two vectors are called orthogonal if . A set of vectors orthogonal to one another is neccesarily linearly independent. The proof is left to the reader as an excercise.
A linearly independent subset of is called a basis set if all vectors in have a unique linear expansion in terms of the basis vectors.
Example 1: Let be the set of all square-integrable functions on the real line segment , and let be any such function. Then, since has a unique Fourier expansion, the set is a basis set for .
Hilbert spaces are frequently taken to be function spaces, that is, spaces whose elements are functions of some kind. The kind of Hilbert space we will be using is called a rigged Hilbert space, in which we generalize to spaces of distributions. In effect, this allows the use of the Dirac delta function, defined by
for any real .
Since a function is given uniquely by specifying its value at all elements of its domain, the set is a basis for any function space on the real line.
Associated with every Hilbert space is the corresponding dual space , consisting of linear functionals on . A linear functional on is a linear function such that .
Let be an orthogonal basis set in . We then construct the set in by sending to , where is the functional given by for all .
Let be an operator on a Hilbert space and concider the equation
.
This is called an eigenvalue equation. is called an eigenvalue of , and an eigenvector. We assume the reader to be familar with the eigenvalue problem in the finite-dimensional case. We will now prove a very useful theorem. If is Hermitian, then the eigenvectors of constitute a basis for .