6—Vector Spaces 156
on the domain−L < x <+Lis just an example of Fourier series, and the components offin this basis are
Fourier coefficientsa 1 ,...,b 0 ,.... An equally valid and more succinctly stated basis is
enπix/L, n= 0, ± 1 , ± 2 , ...
Chapter 5 on Fourier series shows many other choices of bases, all orthogonal, but not usually orthonormal.
6.8 Gram-Schmidt Orthogonalization
From a basis that is not orthonormal, it is possible to construct one that is. This device is called the Gram-Schmidt
procedure. Suppose that a basis is known (finite or infinite),~v 1 , ~v 2 ,...
Step 1: Normalize~v 1 : ~e 1 =~v 1
/√〈
~v 1 ,~v 1
〉
.
Step 2: Construct a linear combination of~v 1 and~v 2 that is orthogonal to~v 1 :
Let~e 20 =~v 2 −~e 1
〈
~e 1 ,~v 2
〉
and then normalize it.
~e 2 =~e 20
/〈
~e 20 ,~e 20
〉 1 / 2
.
Step 3: Let~e 30 =~v 3 −~e 1
〈
~e 1 ,~v 3
〉
−~e 2
〈
~e 2 ,~v 3
〉
etc.repeating step 2.
What does this look like? See problem 3.
6.9 Cauchy-Schwartz inequality
For common three-dimensional vector geometry, it is obvious that for any real angle,cos^2 θ≤ 1. In terms of a
dot product, this is|A~.B~|≤AB. This can be generalized to any scalar product on any vector space:
∣
∣
〈
~u,~v
〉∣
∣≤‖~u‖‖~v‖. (14)
The proof starts from a simple but not-so-obvious point. The scalar product of a vector with itself is by definition
positive, so for any two vectors~uand~vyou have the inequality
〈
~u−λ~v,~u−λ~v
〉
≥ 0. (15)
whereλis any complex number. This expands to
〈
~u,~u
〉
+|λ|^2
〈
~v,~v
〉
−λ
〈
~u,~v
〉
−λ*
〈
~v,~u