Lecture Notes W2

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Class 2: General Linear Spaces

The Complex Vector Space CN

We can generalize the notion of a vector space RN , where the vector elements are
real numbers in R, to CN - where they are complex numbers. One important differ-
ence is that the dual vector will be defined with complex conjugation. To remind,
complex numbers can be defined by their real and imaginary parts,

z = x + iy, x = <(z), y = =(z). (1)

For any complex number, its complex conjugate is defined by z ∗ = x − iy, with

(z ∗ )∗ = z (z1 z2 )∗ = z1∗ z2∗ zz ∗ = z ∗ z = |z|2 = x2 + y 2 .


 
v1
 
 v2  h i
In CN , |vi =   ,
  ∗ ∗
hv| = v1 v2 ... vN . ∗ (2)
 ... 
vN
The inner product is therefore
N
X
hv|ui = vi∗ ui , (3)
i=1

and instead of being symmetric, it follows from the conjugation properties

hv|ui = hu|vi∗ .

The norm of a complex vector is the sum of the norms of its elements
N
X N
X
2
hv|vi = |vi | = < vi 2 + = vi 2 . (4)
i=1 i=1

For a complex matrix A, we define the notion of an adjoint matrix A† by trans-


pose and conjugation:

A† = (A∗ )> or element-wise Aij = A∗ji .

We know that real symmetric matrices, where A> = A, are special - they have a
basis of eigenvectors and other nice properties. Similarly for self-adjoint complex
matrices, where A† = A, they have a basis of eigenvector, and moreover they have
real eigenvalues. We will return to these important operators later.

1
More abstract linear spaces

Imagine an abstract linear space V over a field Φ, where addition and multiplica-
tion by a scalar from the field is defined. Using these operations we can form linear
combinations of the abstract elements. Linear independence follows from the gen-
eralized concept of a linear combination: if a set of elements cannot be combined
to zero with nonzero coefficients, the set is linearly independent.

Example 1: The space of all polynomials of degree N with real coefficients.

a0 + a1 x + a2 x 2 + . . . + aN x N .

Addition is defined by adding the polynomials - that still remains in the space,
and the same for multiplication by a product. In this linear space |0i is the zero
polynomial. The following set

{|v1 i, |v2 i}{|xi, |x2 i}

is linearly independent. This is because one cannot find a linear combination a1 x2 +


a2 x3 = |0i unless a1 = a2 = 0. Note that the zero element in the space is the
polynomial zero, not the nubmer zero.
By the same argument, the set {x, x2 , x3 , ...xN } is a linearly independent set.
However, we cannot find more independent elements, since any other polynomial
of degree N can be created from a combination of this group.
We conclude that this is a space of dimenion N ; in fact it can be mapped onto
RN , by taking the real coefficients of each polynomial and putting them in a col-
umn vector. You may easily convince yourself that these two linear spaces are
equivalent.

Example 2: Space of all polynomials on [−1, 1], of arbitrary finite order. The oper-
ations (addition, multiplication by a scalar) are the same as before; however, this
space is no longer equivalent to RN because of the variable length of coefficient
vectors. The set {x, x2 , x3 ...} is an independent set, showing that we can find an
infinite set of independent elements! It is a basis of the space since any polynomial
can be created by a linear combination of its elements. Therefore, the dimension-
ality of this space is infinite (countable).

Example 3: Space of all matrices of size n × n with real entries. Their addtion is
defined elementwise; multiplication by a scalar - all elements.

2
Example 4: Space of all real functions f (x) on [0, 1].

|f i = f (x).

Addition and multiplication defined as:

(f + g)(x) = f (x) + g(x) (af )(x) = af (x)

|0i is the zero function and the inverse of f is −f .

00
Example 5: The space of solutions of the differential equation y + y = 0 on the
real line. This is a linear space, since for any two solutions y1 (x), y2 (x), their linear
combination a1 y1 (x) + a2 y2 (x) is also a solution. The elements in this space are
also functions, as in the previous example; but it is more constrained since it only
contains solutions.
Consider the two solutions

y1 (x) = cos x, y2 (x) = sin x

are they linearly independent? let us try to construct a zero linear combination

a1 cos x + a2 sin x = 0
p
a1 cos x + a2 (1 − cos2 x) = 0

a21 cos2 x = −a22 (1 − cos2 x) = 0 → (a21 − a22 ) cos2 x = −a22

and this last equation can only be satisfied if a1 = a2 = 0. Therefore, these are two
linearly independent solutions.
Any solution of the differential equation can be written as a combination of
these two, therefore they span the entire space - which means it has dimension 2

3
Inner Product

In a generalized linear space an inner product can be defined that obeys the four
properties (see last class notes). As in RN , the inner product is not unique; in most
cases many alternative definitions are legitimate. A linear space with a particular
inner product is called an ”inner product space”, and is identified by specifying its
elements, and its operations - addition, multiplication and inner product.

Example 1: In the space of real functions f (x), x ∈ [a, b], the following inner prod-
uct can be defined:
Z b
hf |gi = f (x)g(x) dx
a

.
Checking for the inner product properties:
Rb Rb
• hf |gi = a
f (x)g(x) dx = a g(x)f (x) dx = hg|f i
Rb
• hf1 + f2 |gi = a [f1 (x) + f2 (x)]g(x) dx = hf1 | gi + hf2 |gi
Rb Rb
• hcf |gi = a cf (x)g(x) dx = c a f (x)g(x) dx = chf |gi
Rb
• hf |f i = a f 2 (x) dx ≥ 0.

But another requirement regarding the norm is that

hf |f i ↔ f (x) = |0i.

Is this really the case? Suppose a function f (x) is defined to be zero everywhere
except at one point, where it is 7. Its integral is still 0!. We find ourselves forced
to confine the discussion to “wel behaved” functions, or smooth enough. For our
purposes we will most often talk about the space of piecewise continuous func-
tions. These are functions that are continuous everywhere except possibly in a
finite number of points.
Example 2: In the same space of functions of the previous example, we can
define
Z b
hf |gi = f (x)g(x)w(x) dx.
a

where w(x) > 0 is a weight function. Similar to the weighted inner product we
defined in R2 , also here we modify the inner product be counting different points

4
x with different weight. You may easily convince yourself that all inner product
properties are obeyed.

Example 3: In the same way we generalized RN to CN , we can generalize the space


of real functions to a space of complex-valued functions. The dual element of the
complex function |f i = f (x) is then hf | = f ∗ (x), and the analog of the integral
inner product is
Z b
hf |gi = f ∗ (x)g(x) dx.
a

We have a modified set of rules for the inner product over C:

• hv|ui = hu|vi∗

• hv + w|ui = hv | ui + hw|ui

• hv|aui = a hv|ui, hav|ui = a∗ hv|ui

• ||v||2 = hv|vi ≥ 0, with hv|vi = 0 ⇐⇒ |vi = |0i

where we note that in the third property, the notation hav| stands for the dual
element
" # to |avi and therefore
" # includes conjugation of a. For example, if |vi =
1 2+i h i h i
, a = 2+i, |avi = , hav| = 2−i 4−2i = a∗ 1 2 .
2 4+2i

Computing Projections

Since inner products and norms are defined most generally, the formula for pro-
jection holds true in any linear space:
|vihv|
|ui = |u|| i + |u⊥ i = |ui + |u⊥ i.
hv|vi

his is because we derived this formula without actually using a specific form of
the inner product, only with general notation hv|ui. You may go back and check
in lecture notes 1, and appreciate the strength of this general notation without
reference to the details of any particular product. We now illustrate how it works
in some general cases.

5
Example 1: In the space of functions on [0, 1], decompose the linear space element
|gi = 1 − x to the parallel and perpendicular components with respect to |f i = x2 .
Z 1 Z 1 Z 1 1
3 4
x x 1 1 1
hf |gi = x2 (1 − x)dx = x2 dx − x3 dx = − = − =

0 0 0 3 4 3 4 12
0

1
x5
Z
1
hf |f i = x4 dx = =
0 5 5
|gi = |g|| i + |g⊥ i = λ|f i + |g⊥ i
hf |gi 1/12 5 5 2
λ= = = → |g|| i = x
hf |f i 1/5 12 12
5 2
|g⊥ i = |gi − |g|| i = 1 − x −
x.
12
You may check that the two parts add up to the whole function |gi.

Example 2: In the space of real functions on [−π, π] with the integral inner product,
compute the product of |f i = 1 with |gi = sin x.
It is useful to note that the integral of a function which is asymmetric around
some point, over an interval which is symmetric around this point, is always zero.
For example, if g(−x) = −g(x) the function is asymmetric around zero. Then,
Z π
g(x) = 0
−π
RA
as is its integral over any symmetric region: −A
g(x) = 0. All the area you gain
by integrating over the negative part of the interval, is exactly cancelled by its
negative in the positive part.

Example 3: In the space of real functions on [0, π], decompose |gi = sin x to its
projection and orthogonal parts with respect to |f i = 1.
This example looks very much like the previous one, but it is very different! this
is because the space itself is linear: the functions are defined on a different interval.

Example 4: In the same space as the previous example, check the orthonormality
of the following pair of elements:

|f i = sin x, |gi = cos x.

f Example 5: In the space of real polynomials on [−1, 1] and integral inner prod-

6
uct, turn the independent group {1, x, x2 } into an orthonormal group by repeated
projection subtraction and normalization (Gram Schmidt procedure).

Cauchy-Schwartz inequality

In the familiar vector space of RN , the standard inner product has a geometric
interpretation:

hv|ui = ||v||||u|| cos θ

where θ is the angle between the two vectors. (In R2 this is easy to show by writ-
ing the components v1 = |v| cos θv , v2 = |v| sin θv where θv is the angle the vector
v makes with the x-axis, and similarly for u, and multiplying the components ex-
plicitly) Since | cos θ| ≤ 1, it follows that for any two vectors

|hv|ui| ≤ ||v||||u||.

This important formula is known as the Cauchy-Schwartz inequality. The interesting


thing about it is that it holds for ANY two elements in a linear space with an inner
product defined on it.

Example: The following formula is true for any two functions f (x), g(x) on the
interval [0, 1]:

!2
Z 1 Z 1  Z 1 
2 2
f (x)g(x)dx ≤ f (x) dx g(x) dx .
0 0 0

To prove this as a theorem in calculus is quite difficult. But if we remember that


the left hand side is a well-defined inner product, while the right is a product of
two norms, then the inequality is nothing but one implementation of the Cauchy-
Schwartz inequality.

7
Linear Operators

In generalized linear spaces, many linear operators can be defined. Not always can
they be written as matrices; for example, in linear spaces of infinite dimensionality,
there are a infinite number of basis elements, therefore there is no representation as
a finite matrix. We will give a few examples of operatorsin linear function spaces.

Example 1: in the space V of polynomials with real coefficients on the real line,
x ∈ (−∞, ∞), we define the derivative operator, D1 : V → V , as
dp
D1 [p] = D1 p(x) = .
dx
This is an operator in the space, because the derivative of a polynomial is also a
polynomial, and therefore the image remains in the space. Its linear properties
stem from the linearity of the derivative itself:

• D1 |0i = |0i

• D1 [p1 (x) + p2 (x)] = D1 [p1 (x)] + D1 [p2 (x)]

• D1 [cp(x)] = cD1 [p(x)]

The kernel of this operator is all constant polynomials - i.e., polynomials of degree
0. It is a singular operator that cannot be inverted (recall that in calculus, when
“inverting” the derivative by integration, there is always an unspecified constant
that can be added! this shows that there is no well-defined operator to do the
inverse of a derivative).
The derivative operator has no eigenfunctions or eigenvalues (except |0i). When-
ver you differentiate a polynomial, you get a polynomial of a lower degree, which
is linearly independent of the original polynomial and therefore can never be a
scalar multiple of it.

Example 2: in the space V of real differentiable functions on the real line, x ∈


(−∞, ∞), we define the derivative operator, D1 : V → V , as
df
D1 |f i = D1 f (x) = .
dx
This seems the same operator, but it acts in a different space. In this space there are
eigenfunctions and eigenvalues to the operator: the function |fa (x)i = eax ,
d ax
D1 |fa (x)i = a|fa (x)i [e ] = aeax .
dx

8
This operator has infinitely many eigenvectors - a continuum labeled by a.

Example 3: In the same space, we define

d2 f
D2 |f i = D2 f (x) = .
dx2
It is easy to see that this is a linear operator that obeys all requirements - actually,
all orders of derivatives are. All functions of the form

f (x) = A sin kx, k∈R

are eigenvectors of this operator with negative eigenvalues λ = −k 2 . This is again


a continuum. A different continuous set can be defined by

f (x) = Aekx , k∈R

Here all eigenvalues λ = k 2 are positive.

Adjoint and self-adjoint operators

We defined the adjoint operator earlier through a matrix representation. More


generally, any linear operator M has an adjoint operator M† , if for any two vectors
in the space , |ui the following equality holds:

hMv|ui = hv|M† ui.

You may get some intuition for this condition recalling that matrix element Aij was
found by applying the operator on one basis element |ej i and projecting the result
on |ei i. The properties of the adjoint operator are very similar to those of a complex
conjugate:

(M1 + M2 )† = M†1 + M†2


(aM)† = a∗ M†
(M1 M2 )† = M†2 M†1
(M† )† = M.

The importance of self-adjoint operators is that they, like symmetric matrices,


have a basis of

You might also like